This paper presents two new expressions for estimating registration accuracy in point-based rigid-body registration and points out a danger in using the traditional measure of registration accuracy. In this work we present two major results: we first derive an approximate expression for (TRE²), and we then use this to calculate the expected squared alignment error for individual fiducials. This leads to a surprising conclusion--the registration accuracy is generally worst near the fiducials which are most closely aligned. This should demonstrate the danger of relying on the traditional accuracy measure, namely fiducial registration error, to judge the quality of a registration.

In this contribution, we are concerned with the detection and refined localization of 3D point landmarks. We propose multi-step differential procedures which are generalizations of an existing two-step procedure for subpixel localization of 2D point landmarks. This two-step procedure combines landmark detection by applying a differential operator with refined localization through a differential edge intersection approach. In this paper, we theoretically analyze the localization performance of this procedure for analytical models of a Gaussian blurred L-corner as well as a Gaussian blurred ellipse. By varying the model parameters differently tapered and curved structures are represented. The results motivate the use of an analogous procedure for 3D point landmark localization. We generalize the edge intersection approach to 3D and, by combining it with 3D differential operators for landmark detection, we propose three multi-step procedures for subvoxel localization of 3D point landmarks. The multi-step procedures are experimentally tested for 3D synthetic images and 3D MR images of the human head. We show that the multi-step procedures significantly improves the localization accuracy in comparison to applying a 3D detection operator alone.

During an interventional neuroradiology exam, knowing the exact location of the catheter tip with respect to the patient can dramatically help the physician. An image registration between digital subtracted angiography (DSA) images and a volumic pre-operative image (magnetic resonance or computed tomography volumes) is a way to infer this important information. This mono-patient multimodality matching can be reduced to finding the projection matrix that transforms any voxel of the volume onto the DSA image plane. This modelization is unfortunately not valid in the case of distorted images, which is the case for DSA images. A classical angiography room can now generate 3D X-ray angiography volumes (3DXA). Since the DSA images are obtained with the same machine, it should be possible to deduce the projection matrix from the sensor data indicating the current machine position. We propose an interpolation scheme, associated to a pre-operative calibration of the machine that allows us to correct the distortions in the image at any position used during the exam with a precision of one pixel. Thereafter, we describe some calibration procedures and an associated model of the machine that can provide us with a projection matrix at any position of the machine. Thus, we obtain a machine-based 2D DSA/3DXA registration. The misregistration error can be limited to 2.5 mm if the patient is well centered within the system. This error is confirmed by a validation on a phantom of the vascular tree. This validation also yields that the residual error is a translation in the 3D space. As a consequence, the registration method presented in this paper can be used as an initial guess to an iterative refining algorithm.

Longitudinal magnetic resonance spectroscopy (MRS) studies require accurate repositioning of the volume of interest (VOI) over which measurements are made. In this work we present and evaluate a method for the image-guided repositioning of brain volumes of interest. The point-based registration technique we developed allows the repositioning to be performed on-line (i.e., while the patient is in the scanner). MR image volumes were acquired from six subjects, three scans each over the course of a month. During the first scan, two spectroscopy VOIs are visually selected: one in the frontal white matter, the other in the superior cerebellar vermis. The coordinates of 13 internal brain landmarks were also identified. During both subsequent scans, the same 13 landmarks were also identified, and the transformation that registers the first set of landmarks to the subsequent set is compared. This result is used to automatically map the position of the spectroscopy. VOIs from the first volume to the current volume. For the six subjects evaluated to date, we show an average repositioning error of the spectroscopy VOIs in the order of 1 mm. This accuracy allows us to conclude that any variation in the MR spectra are unlikely to be due to repositioning error.

Inter-subject coregistration is a powerful neuroimaging technique that enables comparison and detection of morphological differences across groups of subjects. The present study uses digital phantoms to evaluate errors in two widely employed approaches to inter-subject coregistration of structural MR images of the brain: the manual step-wise approach and the automated method provided with the software package SPM96. Phantoms were constructed by deforming a high resolution T1-weighted MR image in which we have embedded 12 landmarks. For the manual method the accuracy ranged from 0.8 mm in quadrigeminal plate to 2.4 mm in superior central sulcus and occipital lobe. The average error was 1.5 mm. For the automated SPM96 method and the 9 parameter model, the accuracy ranged from 0.8 mm to 2.1 mm and averaged 1.1 mm. Error of the manual method correlated strongly with the distance from the center of the image (r equals 0.77, slope equals .020, p equals .003). The linear correlation of the error obtained with the automated method with the distance was poor (r equals 0.39, slope equals .008, p > 0.2). The results suggest that the inferior performance of the manual method is due to its step-wise approach and to a relatively large rotational error.

We recently reported a hybrid registration technique that uses a weighted combination of multiple geometrical feature shapes. In this study we use the weighted geometrical feature (WGF) algorithm to register CT images of the head to physical space using the skin surface only, the bone surface only, and various weighted combinations of these surfaces and one fiducial point (centroid of a bone-implanted marker). We use data acquired from six patients that underwent temporal lobe craniotomies for the resection of cerebral lesions. Each patient had four external markers attached to transcutaneous posts screwed into the outer table of the skull. We evaluate and compare the accuracy of the registrations obtained using these various approaches by using as a gold standard the registration obtained using three of the four bone-implanted markers (the remaining marker is used in the various combinations). The results demonstrate that a combination of geometrical features can improve the accuracy of CT-to-physical space registration. The WGF algorithm might thus be useful in various image- guided surgical systems.

We describe a new method for rigid registration of a pre- operative CT-scan image to a set of intra-operative x-ray fluoroscopic images, for guiding a surgical robot to its trajectory planned from CT. Our goal is to perform the registration, i.e. compute a rotation and translation of one data set with respect to the other to within a prescribed accuracy, based upon bony anatomy only, without external fiducial markers.

More than 250,000 lumbar spin operations are performed annually in the United States. Because of this large number of projected surgeries, improvements in surgical procedures can have a tremendous practical impact. We are developing a method to quickly and accurately register surface points extracted from ultrasound images of a vertebra to surfaces extracted from computed tomography images. Identification of surface points in ultrasound images is a challenging task, since points must be extracted in real time and sine large portions of the vertebral surface are invisible in the ultrasound image.

In this study we have examined errors in measurements of vertebral heights and vertebral area resulting from spin rotation and projection effects in x-ray images. Measurement errors were evaluated with phantom images, and simulated rotations of a 3D spine model. An active contour model (snake) was used for measurements of vertebral area. The model contained two pressure parameters which were needed to obtain good fits of the snake to upper and lower edges (endplates) of rotated vertebral bodies. Details of the snake model are included in this report. The results of this study indicate that six point vertebral morphometry can result to significant measurement errors, representing an overestimation of vertebral height and area, in cases showing projection effects and concealed endplate contours. In serial studies, such errors could produce the erroneous appearance of `growing' vertebral bodies. One can improve the accuracy of the morphometric analysis by using additional fiducial points placed on corresponding endplate contours. Additional useful information on fracture and vertebral deformity can be obtained by accurately tracking edge contours, using an active contour model, or comparable techniques.

A reliable and accurate registration technique is a critical requirement for computerized analysis of temporal changes of suspicious lesions in mammograms. Temporal mammographic changes manifest as `neodensities' and may, sometimes, be the only finding carcinoma. Because of the elasticity of the breast tissue, the absence of obvious landmarks, and the large variability in the relative position of the breast tissue projected on to the mammogram from one examination to the other, conventional registration techniques are not applicable to registration of breast images. We are developing an automated regional registration technique to simulate the method used by radiologists for identifying corresponding lesions on mammograms.

Automated multi-modality 3D medical image alignment has been an active area of research for many years. There have been a number of recent papers proposing and investigating the use of entropy derived measures of brain image alignment. Any registration measure must allow us to choose between transformation estimates based on the similarity of images within their volume of overlap. Since 3D medical images often have a limited extent and overlap, the similarity measure for the two transformation estimates may be derived from two very different regions within the images. Direct measures of information such as the joint entropy and mutual information will therefore be a function of, not only image similarity in the region of overlap, but also of the local image content within the overlap. In this paper we present a new measure, normalized mutual information, which is simply the ratio of the sum of the marginal entropies and the joint entropy. The effect of changing overlap on current entropy measures and this normalized measure are compared using a simple image model and experiments on clinical MR-PET and MR-CT image data. Results indicate that the normalized entropy measure provides significantly improved behavior over a range of imaged fields of view.

Recent studies indicate that maximizing the mutual information of the joint histogram of two images is an accurate and robust way to rigidly register two mono- or multimodal images. Using mutual information for registration directly in a local manner is often not admissible owing to the weakened statistical power of the local histogram compared to a global one. We propose to use a global joint histogram based on optimized mutual information combined with a local registration measure to enable local elastic registration.

Previous image registration schemes based on mutual information use Shannon's entropy measure, and they have been successfully applied for mono- and multimodality registration. There are cases, however, where maximization of mutual information does not lead to the correct spatial alignment of a pair of images. Some failures are due to the presence of local or spurious global maxima. In this paper we explore whether the normalization of mutual information via the use of a weight based on the size of region of overlap, improves the rate of successful alignments by reducing the presence of suboptimal extrema. In addition, we examine the utility of a deterministic entropy measure. The results of the present study indicate that: (1) the normalized mutual information provides a larger capture range and is more robust, with respect to optimization parameters, than the non-normalized mutual information, and (2) the optimization of mutual information with the deterministic entropy measure takes, on average, fewer iterations than when using Shannon's entropy measure. We conclude that the normalized mutual information using the deterministic entropy measure is a faster and more robust function for registration than the traditional mutual information.

Although a large variety of image registration methods have been described in the literature, only a few approaches have attempted to address the rigid registration of medical images showing gross dissimilarities (due for instance to lesion evolution). In the present paper, we develop driven registration algorithms, relying on robust pixel similarity metrics, that enable an accurate (subvoxel) rigid registration of dissimilar single or multimodal 2D/3D images. In the proposed approach, gross dissimilarities are handled by considering similarity measures related to robust M-estimators. A `soft redescending' estimator (the Geman- McClure p-function) has been adopted to reject gross image dissimilarities during the registration. The registration parameters are estimated using a top down stochastic multigrid relaxation algorithm. Thanks to the stochastic multigrid strategy, the registration is not affected by local minima in the objective function and a manual initialization near the optimal solution is not necessary. The proposed robust similarity metrics compare favorably to the most popular standard similarity metrics, on patient image pairs showing gross dissimilarities. Two case studies are considered: the registration of MR/MR and MR/SPECT image volumes of patients suffering from multiple sclerosis and epilepsy.

MRI has been proposed as an alternative method to mammography for detecting and staging breast cancer. Recent studies have shown that architectural features of breast masses may be useful in improving specificity. Since fractal dimension (fd) has been correlated with roughness, and border roughness is an indicator of malignancy, the fd of the mass border is a promising architectural feature for achieving improved specificity. Previous methods of estimating the fd of the mass border have been unreliable because of limited data or overlay restrictive assumptions of the fractal model. We present preliminary results of a statistical approach in which a sample space of fd estimates is generated from a family of self-affine fractal models. The fd of the mass border is then estimated from the statistics of the sample space.

Texture is an important attribute which is widely used in various image analysis applications. Among texture features, morphological texture features are least utilized in medical image analysis. From a computational standpoint, extracting morphological texture features from an image is a challenging task. The computational problem is made even greater in medical imaging applications where large images such as mammograms are to be analyzed. This paper discusses an efficient method to compute morphological texture features for any geometry of a structuring element corresponding to a texture type. A benchmarking of the code on three machines (Sun SPARC 20, Pentium II based Dell 400 workstation, and SGI Power Challenge 10000XL) as well as a parallel processing implementation was performed to obtain an optimum processing configuration. A sample processed mammogram is shown to illustrate the code outcome.

This paper describes a method of distinguishing between early malignant melanoma and benign moles by examining skin pattern texture on an image of the lesion. Skin patterning is a macroscopic texture composed of fine linear elements. This texture is poorly described by standard definitions of texture and poorly detected by existing techniques. Skin line patterning is detected through a new method which looks at small patches spaced equally across the image and constructs a profile of their linear self-similarity over a range of angles. Regions which exhibit skin patterning result in similar profiles for neighboring patches whereas no such similarity is found in areas where the patterning is disrupted. Interpretation of the profile images for the classification of the lesions is then addressed.

Various researchers have used texture analysis to perform ultrasound tissue characterization with mixed results. Others have used spectral parameters to classify tissue. Several groups have used a feature attributed to the mean scatterer spacing (MSS) as a discriminating feature for tissue classification. We have previously shown that the locations of peaks in the complex cepstrum (PCEP) provide a good estimate of MSS in phantom and simulation studies. In this work, we show that by combining PCEP with texture-based features, we are able to distinguish between normal and hepatitis patients with a high degree of accuracy.

Texture feature extraction is a fundamental stage in texture analysis. Therefore, the reduction of its computational time and memory requirements should be an aim of continuous research. The Spatial Gray Level Dependence Method (SGLDM) is one of the most significant statistical texture description methods, especially in medical image analysis. However, the co-occurrence matrix is inefficient in terms of time and memory requirements. This is due to its dependency on the number of grey levels in the entire image. Its inefficiency puts up barriers to the wider utilization of the SGLDM in a real application environment. This paper investigates the space and time efficiency of self-adjusting binary search trees, in replacing the co-occurrence matrix. These dynamic data structures store only the significant textural information extracted from an image region by the SGLDM. Furthermore, they have the ability to restructure themselves in order to adapt to the co-occurrence distribution of the grey levels in the analyzed region. This results in a better time performance for texture feature extraction. The proposed approach is applied to a number of magnetic resonance images of the human brain and the human femur. A comparison with the co-occurrence matrix, in terms of space and computational time, is performed.

The objective of this research was to develop a method for measuring small differences in diagnostic task performance between two imaging modalities. In the proposed Differential Receiver Operating Characteristic (DROC) method, the observer is shown a pair of images of the same patients, one from each modality A and B. The patient can be normal or abnormal but this information is not known to the observer. The observer selects the image that is preferred for the specified diagnostic task and assigns a rating. Analysis of this experiment yields the area under the DROC curve (A_d). If A_d is greater than 0.5, then modality B is superior to modality A and conversely, if it is less than 0.5, then A is superior to B. With a plausible assumption it can be shown that the result of the paired image presentation DROC experiment will track the difference in A_z's obtained from two single image presentation ROC experiments. A proof-of-concept experiment was conducted to test this idea. Both DROC and ROC studies were conducted using 6 readers. The DROC method was found to track the conventional ROC method and to yield far greater sensitivity, on the average about a factor of 3.46 greater. This work suggests a new paradigm for differential observer performance experiments. The DROC method has the potential for detecting extremely suitable differences in image quality, much smaller than is detectable by the present ROC method. This should allow rapid optimization of imaging systems, without the need for expensive and often inconclusive ROC studies.

A decision model is presented to increase the specificity of breast biopsy directly optimized on the receiver operating characteristic (ROC) area index. ROC area has higher clinical significance as a performance measure than the traditional metric mean-squared error (MSE). Excisional biopsy as practiced is highly sensitive to cancer but nonspecific; only one in three biopsies is malignant. Data for this study consists of 500 cases randomly selected from patients who underwent excisional biopsy for definitive diagnosis of breast cancer. For each case, inputs to the model consist of mammographic findings and patient history features. Outputs from the model built may be thresholded to correspond to the decision to biopsy a suspicious breast lesion. While clinically relevant, ROC area is a discontinuous function which cannot be optimized directly so a genetic algorithm approach is used to train a nonlinear artificial neural network. Performance using the genetic algorithm method of training was similar to that of a decision model trained using the traditional approach for this data set. ROC areas were obtained after training using three different approaches: genetic algorithm training optimized on ROC area produced an ROC area of 0.845 +/- 0.039, genetic algorithm training optimized on MSE produced an ROC area of 0.845 +/- 0.039, and traditional training using backpropagation produced an ROC area of 0.848 +/- 0.039. Despite the similar performance measures for models trained on this data, it is possible that with different data sets, training on ROC instead of MSE will produce models with significantly different performance. In this case, the genetic algorithm approach will prove useful.

Detection of abnormalities is critical to the success of mammogram screening and represents a perceptual problem even for experienced radiologists. This perceptual problem makes the development of reliable automated methods for detection of abnormalities very important. The present work demonstrates improvements in the existing techniques for detection of masses by using an evidential approach to mammogram segmentation. A method of partitioning mammograms into homogeneous regions by using `generic' label is presented. This method assigns the same label to regions based on similarity between regions in the feature space and does not require estimation of model parameters from specific region samples. The features best suited to represent the difference between tissue and masses texture are selected and combined within the framework of the Dempster-Shafer Theory of Evidence. Utilization of the Dempster-Shafer Theory of Evidence has improved the accuracy of detection by allowing to incorporate any number of different features and deal with the uncertainty inherent in these types of problems. Exploitation of constraints, representing domain knowledge, to forbid certain configuration of regions during segmentation results in an improved partitioning of the mammograms. A constrained stochastic relaxation algorithm is used for building an optimal label map to separate tissue and masses.

The fidelity of the reconstructed image in an image coding/decoding scheme and the lowest transmission bit rate from rate-distortion theory can be predicted provided the image statistics are known. Currently popular subband image coding assumes Gaussian source with memory for optimal performance. However, most images do not follow the ideal distribution. The advantage of subband coding lies in the fact that the wavelet coefficients in decomposed subimages have probability distribution functions (pdf's) that can be modeled as a generalized Gaussian when proper parameters are chosen experimentally. However, the filter length chosen for digital implementation of a specific wavelet is crucial in shaping the pdf characteristics and hence in the ability to predict the achievable bit rate at minimum distortion in a quantization scheme. We have analyzed the pdf's of a number of wavelets and chosen filter lengths providing the best fit to a generalized Gaussian distribution for encoding an image by vector quantization of multiresolution wavelet subimages using an adaptive clustering. Our results demonstrate that the performance of the adaptive vector quantizer improves significantly when wavelet filter lengths are chosen to fit the generalized Gaussian distribution.

The classification of brain PET volumes is carried out in three main steps: (1) registration, (2) feature extraction and (3) classification. The PET images were already smoothed with a 16 mm isotropic Gaussian kernel and registered within the Talairach and Tournoux reference system. To make the registration more accurate over a single reference, a method based on optical flow was applied. Feature extraction is carried out by principal component analysis (PCA). Support vector machines (SVM) are then used for classification, because they are better controlled than neural networks (NN) and well adapted to small sample size problems. SVM are constructed by a training algorithm that maximizes the margin between the training vectors and the decision boundary. The algorithm is simple quadratic programming under linear constraints, which leads to global optimum. The decision boundary is expressed as a linear combination of supporting vectors which are a subset of the training vectors closest to the decision boundary. After registration, NN and SVM were trained with the features extracted by PCA from the training set. The estimate error rate is 7.1% for SVM and 14.3% for NN.

In 1996, the University of Iowa launched a multiphase project for the development of a well-structured interdisciplinary image systems engineering curriculum with both depth and breadth in its offerings. This project has been supported by equipment grants from the Hewlett Packard Company. The new teaching approach that we are currently developing is very dissimilar to that we used in previous years. Lectures consist of presentation of concepts, immediately followed by examples, and practical exploratory problems. Six image processing classes have been offered in the new collaborative learning environment during the first two academic years. This paper outlines the employed educational approach we are taking and summarizes our early experience.

More and more often, undergraduate students express the desire to take a course on image processing. These students will learn the most if the theory and algorithms covered in class can be not only illustrated through examples shown by the instructor during class but also coded, tested, and evaluated by the class participants. In the past, the major hurdle to developing a hands-on approach to image processing instruction has been the amount of programming required to implement relatively simple applications. Typical undergraduate students lack experience with low level programming languages and time is spent teaching the language itself rather than experimenting with the algorithms. High level and interpreted programming languages such as Matlab permit to address this question. Even with very little practical exposure to the language, students can rapidly develop the level of skills required to implement a range of image processing algorithms. This presentation will go over the material covered in a senior level introductory course in image processing taught at Vanderbilt University. The course itself is taught in a traditional way but it is supported by laboratories during which students are asked to implement algorithms ranging from connected component labeling to image deblurring. The students are also assigned projects that span several weeks. Examples of such assignments and projects are presented.

This paper gives a comparative overview of ten of the currently available computer vision and image processing textbooks. These texts differ significantly in their coverage, scope, approach, and target audience. Because of the multi-disciplinary nature of this field, it is important to select a textbook that takes advantage of students' backgrounds and gives them the foundation necessary to integrate diverse concepts. This comparative review provides computer vision and image processing educators with a starting point from which they can select a textbook appropriate for their students' needs.

We are developing computerized methods to distinguish between malignant and benign lesions in contrast-enhanced magnetic resonance images of the breast. In this study, we compare 2D spatial analysis of lesions with 3D spatial analysis. Our database consists of 28 lesions: 15 malignant and 13 benign. At 90 s intervals, 4 to 6 scans are obtained, and the spatial uptake of contrast agent is analyzed. Computer-extracted features quantify the inhomogeneity of uptake, sharpness of the margins, and shape of the lesion. Stepwise multiple regression is employed to obtain a subset of features, followed by linear discriminant analysis to estimate the likelihood of malignancy. Cross-validation and ROC analysis are used to evaluate the performance of the method in distinguishing between benign and malignant lesions. The procedures are performed in 3D, and in 2D from single and multiple slices. Shape and sharpness of the lesion were the most effective features. ROC analysis yielded an A_z value of 0.96 for 3D features, between 0.67 and 0.92 for single slices, and 0.88 for 2D features from multiple slices. The performance of 2D analysis on single slices depends strongly on the selected plane and may be significantly lower than the accuracy of full 3D analysis.

Breast ultrasound can potentially increase the accuracy of computerized discrimination of malignant and benign masses. Newly developed 3D ultrasound techniques provide statistically richer information than conventional 2D ultrasound, and may therefore be better-suited for computerized statistical classification techniques. In this study, we investigated the feasibility of classifying solid breast masses using features extracted from 3D ultrasound images. Our data set consisted of seventeen biopsy-proven masses. Eight of the masses were malignant and nine were benign. The masses were identified by an experienced breast radiologist in the 3D volume, and a 3D ellipsoid containing the mass was defined. Spatial gray level dependence features were extracted from 2D slices in three regions, which were (1) the interior of the ellipse; (2) a disk-shaped region at the upper periphery of the ellipse; and (3) a disk-shaped region at the lower periphery of the ellipse. 2D analysis was performed by evaluating the classification accuracy of the features extracted from each slice. 3D analysis was performed by first averaging feature values from different slices into a single 3D feature, and then evaluating the classification accuracy. The best texture feature in this study achieved a classification accuracy of A_z equals 0.97 for both 3D and 2D analysis. Our results indicate that the performance of 3D analysis is comparable to that of 2D analysis using the best available slice. Since the best 2D slice for texture analysis may not be known a-priori, this preliminary study suggests that 3D ultrasound may be beneficial for computerized breast mass characterization.

We are developing computerized schemes to detect clustered microcalcifications in mammograms, and to classify malignant versus benign microcalcifications. The purpose of this study is to investigate the effects on the performance of computer classification when results of computer-detected true microcalcifications and computer detected false-positive signals are used as input to the computer classification scheme. We found that when trained using manually identified microcalcifications, the computer classification performance was not degraded significantly when up to 60% of true microcalcifications were missed, and when false-positive signals made up approximately one half of the computer detection.

A computer-aided diagnosis (CAD) scheme could improve radiologists' diagnostic performance in their detection of lung nodules on chest radiographs if the computer output were used as a `second opinion'. The current CAD scheme that we have developed achieved a performance of 70% sensitivity and 1.7 false positives per image for our database. This database consisted of two hundred PA chest radiographs, including 100 normals and 100 abnormals (containing 122 confirmed nodules). Our purpose in this study was to improve our scheme further by incorporating new features derived from analysis of the histogram of radial edge gradients on nodule candidates.

3D reconstruction of medical images is increasingly being used to diagnose disease and to direct therapy. Virtual bronchoscopy is a recently developed type of 3D reconstruction of the airways that may be useful for diagnosis of lesions of the airway. In this study, we compare two methods for computer-aided diagnosis of polypoid airway tumors: a parametric (`patch') and non-parametric ('grey-scale') algorithm. We found that both methods have comparable specificities. Although the non-parametric method is twelve times faster than the parametric method, we found that is sensitivity lags behind that of the parametric method by 3 to 16% when lesions of all sizes are considered. For lesions at least 5 mm in size, the sensitivities are comparable if a small convolution kernel is used.

One of the basic assumptions in the computed tomography (CT) is that the scanned object has constant attenuation characteristics across the thickness of the slice. In reality, however, this assumption is often violated. The projection data set of a head CT scan, for example, is often corrupted by bony structures partially intruded into the scanning plane. As a result, severe streaking and shading artifacts appear in the reconstructed images. This phenomenon is called partial volume. In this paper, we propose an iterative approach to the partial volume artifact reduction. CT images are first reconstructed with a filtered backprojection algorithm. The generated images subsequently undergo a fuzzy membership classification process to arrive at bone-only images, which in turn will be used to produce gradient images. The projection error is then calculated based on the gradient image. For better error estimation, the scan data is collected in a helical mode and highly overlapped images are reconstructed. The error term is filtered and back-projected to produce a partial volume error image, which is scaled and subtracted from the original image. Various phantom studies have demonstrated the effectiveness of our approach.

In this paper the reconstruction of SPECT images is explored using the Total-Variation method. The method is able to recover edge information of the image and is therefore suitable for `blocky' types of images. The paper provides a basic reconstruction algorithm with examples using synthetic and patient data.

Neural networks are applied to a Rayleigh discrimination task for simulated limited-view computed tomography with maximum-entropy reconstruction. Network performance is compared to that obtained using the best machine approximation to the ideal observer found in an earlier investigation. Results obtained on 2D subimage inputs are compared with those for 1D inputs and presented previously at this conference. Back-propagation neural networks significantly outperform the `best' standard nonadaptive linear machine observer and also the intuitively appealing `matched filter' obtained by averaging over the images in a large training data set. In addition, the back-propagation neural network operating on 2D subimages performs significantly better than that limited to 1D inputs. Finally, improved performance on this Rayleigh task is found for nonlinear (over linear, that is, simple perceptron) neural network decision strategies.

The detection of scattered photons affects both image quality and accuracy of quantitation accuracy in Single Photon Emission Computed Tomography (SPECT). The aim of this work was to evaluate three scatter correction methods: Jaszczak subtraction, the triple energy window method and an artificial neural network based approach. This evaluation was performed not only in terms of contrast and spatial resolution but also in terms of absolute and relative quantitation. A Monte Carlo simulation of an anthropomorphic cardiac phantom allowed us to obtain a realistic SPECT study while knowing the primary (non scattered) photon distribution. The knowledge of the primary activity made possible the study of the effect of scatter alone independently on all other phenomena affecting quantitation. The quantitative error propagation between the projections and the reconstructed slices due to scatter was studied as well as resolution, contrast and uniformity recoveries in the corrected images. The results show that an artificial neural network achieved the best scatter correction both in terms of relative (gives the same uniformity as in the primary distribution) and absolute quantitation (error < 4%) and resolution. The triple energy window method led to good quantitation (error < 8%) and contrast results but poorer resolution recovery than the artificial neural network based approach. Jaszczak subtraction yielded good quantitation (error < 7%) but introduced severe non uniformities in the image (decrease of the uniformity by 35%).

The Markov Chain Monte Carlo (MCMC) technique provides a means to generate a random sequence of model realizations that sample the posterior probability distribution of a Bayesian analysis. That sequence may be used to make inferences about the model uncertainties that derive from measurement uncertainties. This paper presents an approach to improving the efficiency of the Metropolis approach to MCMC by incorporating an approximation to the covariance matrix of the posterior distribution. The covariance matrix is approximated using the update formula from the BFGS quasi-Newton optimization algorithm. Examples are given for uncorrelated and correlated multidimensional Gaussian posterior distributions.

A Maximum Likelihood estimation technique is proposed for optimal estimation of Magnetic Resonance (MR) T₂ maps from a set of magnitude MR images. Thereby, full use is made of the actual probability density function of the magnitude data, which is the Rician distribution. While equal in terms of precision, the proposed method is demonstrated to be superior in terms of accuracy compared to conventional relaxation parameter estimation techniques.

In most clinical MR images, global patient motion is predominantly interview (between readouts), and corrupts the phase of the received signal for that line of k-space. No information, however, is lost--if the original motion is known and appropriate phase corrections applied, the image can be perfectly restored. It is therefore possible to correct for motion given only the real and imaginary data from the scanner by simply trying different possible motion corrections and searching for the highest quality resulting image with a suitable evaluation function. Such an `autofocusing' algorithm was recently described, using image entropy as the cost function; however, very long computation times are required. If the corrupting motion is primarily 1D, much faster autofocusing might be possible by calculating only selected lines of the image. In this paper, we describe work on such an algorithm, implemented with both minimum entropy and maximum variance as the cost functions. Tests on several 256 X 256 magnitude images artificially corrupted by 1D motion indicate that evaluating only eight selected columns of the image (calculated with eight 1D FFT's) works very well--essentially as well as evaluating the whole image, which requires 2D FFT's. The run time dropped from several hours for 2D FFT's to less than ten minutes using 1D FFT's. One test image with little dark area was not well corrected, indicating the possible dependence of both cost functions on dark regions to be cleared of artifacts.

This paper presents a fast method for magnetic resonance imaging (MRI) based on discrete cosine transform (DCT). In the proposed method, DCT is used for phase encoding and discrete Fourier transform (DFT) is used for frequency encoding. Because of the superior information compression property of DCT compared to DFT, the proposed method requires a smaller portion of the k-space to generate acceptable MRI images. Thus, the new method reduces the imaging time and increases the patient throughput. The hardware modifications for generating DCT encoded free induction decay signals and reconstruction algorithms to generate tomographic images are presented. Capability of the method in generating MRI images is illustrated through computer simulations. Finally, the effect of MRI noise on the quality of the resulting images are shown by simulation studies.

When conventional techniques are employed in the quantitative analysis of phase contrast magnitude Magnetic Resonance (MR) data, the results obtained are biased. The bias is due to the contributions from inherent random noise. To remove this bias, knowledge of the actual shape of the data probability density function becomes essential. In the present work, the full knowledge of the probability distribution of the phase contrast magnitude MR data is exploited for optimal estimation of the underlying signal.

A new method for compensating motion artifacts in computerized tomography is presented. The algorithm operates on the raw data, that is, on the measured Radon transform. An edge detection is performed on the Radon transform image. Two curves are obtained from these edges, which are fitted by a polynomial. The pointwise differences between the edges of the Radon transform image and the fitted curves are used to determine the parameters for the motion compensation. The corresponding operations are performed on the rows of the Radon transform. Our method is independent of the image reconstruction algorithms like filtered backprojection, Fourier method, etc. since all transformations are performed on the raw data.

Acoustic shear waves of low frequency can be detected and measured using a phase contrast based magnetic resonance imaging technique called MR Elastography or correlation or phase measurement based echo ultrasound techniques. Spatio- temporal variations of displacements caused by the propagating waves can be used to estimate local values of the elasticity of the object being imaged. The currently employed technique for estimating the elasticity from the wave displacement maps, the local frequency estimator (LFE), has fundamental resolution limits and also has problems with shadowing and other refraction-related artifacts. These problems can be overcome with an inverse approach using Green's function integrals which directly solve the wave equation problem for the propagating wave. The complete measurements of wave displacements as a function of space and time over the object of interest obtained by the above techniques make possible an iterative approach to inversion of the wave equation to obtain elasticity and attenuation maps. The speed of convergence of such an iterative method can be improved by using LFE as the initial guess for the object function. This article describes the proposed and evaluates the improvements over the LFE for simulated data and in-vivo breast measurements.

Iterative methods for the reconstruction of PET images can produce results superior to filtered backprojection since they are able to explicitly model the Poisson statistics of photon pair coincidence detection. However, many implementations of these methods use simple forward and backward projection schemes based either on linear interpolation or on computing the volume of intersection of detection tubes with each voxel. Other important physical system factors, such as depth dependent geometric sensitivity and spatially variant detector pair resolution are often ignored. In this paper, we examine the effect of a more accurate system model on algorithm performance. A second factor that limits the performance of the iterative algorithms is the chosen objective function and the manner in which it is optimized. Here we compare performance of filtered backprojection (FBP) with the OSEM (ordered subsets EM) algorithm, which approximately maximizes the likelihood, and a MAP (maximum a posteriori) method using a Gibbs prior with convex potential functions. Using the contrast recovery coefficient (CRC) as a performance measure, we performed various phantom experiments to investigate how the choice of algorithm and projection matrix affect reconstruction accuracy. Plots of CRC versus background variance were generated by varying cut-off frequency in FBP, subset size and iteration number or post-smoothing kernel in OSEM, and smoothing parameter in the MAP reconstructions. The results of these studies show that all of the iterative methods tested produce superior CRCs than FBP at matched background variance. However, there is also considerable variation in performance within the class of statistical methods depending on the choice of projection matrix and reconstruction algorithm.

This paper describes a new image reconstruction method to reduce the presence of image artifacts in scintimammography. SPECT data are mathematically modified prior to conventional image processing, followed by an inverse transform, which permits the tomographic visualization of low signals in the presence of large background intensities observed in scintimammography. Images of Tc-99m sestamibi distribution were obtained in a modeled `breast tissue', with a high activity in the `myocardium' as compared to the `breast tissue' (i.e. 20:1). Image reconstruction with the modified algorithm were qualitatively correct and demonstrated the regions of enhanced uptake (lesions) with no evidence of artifacts from the background counts in the heart. Comparison with MRI demonstrated that the hot regions were properly located and correlated with the MRI data. In contrast, a standard (i.e. Filtered Back Projection) reconstruction resulted in streak artifacts in place of `breast tissue' which rendered them clinically useless. This new approach to scintimammography offers the prospect of significantly reducing image artifacts and improving imaging accuracy.

The MUSIC (Multiple Signal Classification) algorithm is a recently proposed method in solving multiple dipole localization problem from spatio-temporal magnetocardiograph (MCG) data. There are many factors that may effect the resolvability of MUSIC method in solving MCG inverse problem. For example, the number and space arrangement of sensors, the signal-noise ratio of measurement data, the relative position of dipole to the sensors, the direction of dipole. In the case of multiple dipoles are assumed, the distance and time correlation between the dipoles may take a great effect on the solution accuracy. We need a quantitative method of evaluate the resolvability of MUSIC algorithm. In this paper spherically symmetric conductor model is applied as the forward model. The statistical performance of the MUSIC algorithm is discussed by using the MUSIC error covariance matrix. The Cramer-Rao Lower Bound (CRLB) on localization errors for MCG current source dipole models is presented. The performance of MUSIC algorithm is compared with the ultimate performance corresponding to CRLB. The numerical studies with simulated MCG data are presented in two cases: one dipole is assumed and two dipoles are assumed.

This paper describes a modeller for building 3D anatomic atlases. The main contribution is the ability to incrementally learn inter-patient variations in structure, shape, and function; with support for variations in structure (topology) being of particular significance. The modeller uses graph theory as a mathematical base and is, in fact, quite general with possible applications in wider computer vision contexts. In this paper we describe the general modeller, and illustrate it specifically by showing how to use it to build a catalogue of vasculature and variations. We have used such a catalogue of vasculature for 3D reconstruction of x-ray angiograms, simulation of x-ray angiography, and interpretation of images and text combined; the first two of these are the most mature and are briefly outlined. We conclude by discussing the limitations of the model both generally and specifically.

In this paper we present an approach for 3D reconstruction of vascular skeletons from x-ray angiograms at arbitrary angles. The key to this problem is solving the correspondence problem; that is, determine a mapping between points in each image pair. This paper describes, in some detail, the two main steps in our approach: (1) we determine vessels that correspond between images to obtain a `coarse grained' solution that acts to focus our next step; (2) we determine the correspondence between points in corresponding vessels, this gives a `fine grained' solution enabling the 3D shape of the vessel to be computed. We use a model of a collection of vasculature during the first step, use the second step relies only on general assumptions about space- curves. There are several novelties: images can be at arbitrary angles; generalization from the collection to include new vasculature; and weak dependence on the kind of image data. Results show the approach is efficient in time and space, accurate and reliable, and robust to noise.

Segmentation and 3D reconstruction of vessel center lines from subtraction angiography is difficult because of noise, an uneven distribution of the contrast agent, and bone structures concealing the vessels. In biplane images, some ambiguities in segmentation in one of the images can be resolved using image information from the other image. Images are first processed separately. A-priori vessel probabilities are computed using the grey value in the image. Gradients and second derivatives are computed using gaussian derivatives. Subsequently, 3D reconstruction is carried out in a line-by-line fashion between user-specified base lines on both images. The user is asked to point out corresponding vessel cross-sections on the frontal and lateral image. 3D locations are computed. Between adjacent 3D locations, intermediate 3D locations are estimated iteratively using location, curvature and size of the previously computed 3D vessel cross-section. The estimates are optimized evaluating local attributes in the two projection images, such as the grey level, the gradient and the closeness to the initial estimate. WE applied the method for the reconstruction of artificial and real structures. The artificial object consisted of metal wires with sizes ranging between 1 and 5 mm. Various parts of the structure were extracted and reconstructed successfully. Reconstructing center lines from real images was more difficult because of the image degrading influences. Structures with a size of about 1 mm were extracted and reconstructed successfully even though they were barely visible. The amount of user interaction depended on the visibility of the structures in the two images.

The 3D coronary vessels can be reconstructed by means of different cardiac imaging modalities. Two of the most widely used modalities for the purpose of coronary tree reconstruction are intravascular ultrasounds (IVUS) and biplane angiography. Current 3D vessel reconstruction based on IVUS pullback imaging is limited by the lack of information about the real vessel curvature, because the path of the catheter is assumed to be a straight line. This limitation can be overcome if information from an IVUS sequence is fused with a biplane X-ray image of the catheter acquired at the start of the pullback procedure. This work focuses on the reconstruction of the catheter path from biplane angiograms. This reconstruction represents the 3D path followed by the catheter inside the vessel of interest. While other approaches reconstruct the vessel after it has been segmented in both images independently, our approach, based on the snakes technique, allows us to segment and reconstruct the catheter trajectory merging information from both images simultaneously. The result is a more robust reconstruction since 3D constraints can be used and no correspondence of points between the projections is required. This reconstruction will allow a posterior more exact combination of IVUS and biplane angiography image modalities.

X-ray mammography is one of the most significant diagnosis methods in early detection of breast cancer. Usually two X- ray images from different angles are taken from each mamma to make even overlapping structures visible. X-ray mammography has a very high spatial resolution and can show microcalcifications of 50 - 200 micron in size. Clusters of microcalcifications are one of the most important and often the only indicator for malignant tumors. These calcifications are in some cases extremely difficult to detect. Computer assisted diagnosis of digitized mammograms may improve detection and interpretation of microcalcifications and cause more reliable diagnostic findings. We build a low-cost mammography workstation to detect and classify clusters of microcalcifications and tissue densities automatically. New in this approach is the estimation of the 3D formation of segmented microcalcifications and its visualization which will put additional diagnostic information at the radiologists disposal. The real problem using only two or three projections for reconstruction is the big loss of volume information. Therefore the arrangement of a cluster is estimated using only the positions of segmented microcalcifications. The arrangement of microcalcifications is visualized to the physician by rotating.

This paper presents reproducibility studies for the segmentation results obtained by our optimal MRI feature space method. The steps of the work accomplished are as follows. (1) Eleven patients with brain tumors were imaged by a 1.5 T General Electric Signa MRI System. Four T2- weighted and two T1-weighted images (before and after Gadolinium injection) were acquired for each patient. (2) Images of a slice through the center of the tumor were selected for processing. (3) Patient information was removed from the image headers and new names (unrecognizable by the image analysts) were given to the images. These images were blindly analyzed by the image analysts. (4) Segmentation results obtained by the two image analysts at two time points were compared to assess the reproducibility of the segmentation method. For each tissue segmented in each patient study, a comparison was done by kappa statistics and a similarity measure (an approximation of kappa statistics used by other researchers), to evaluate the number of pixels that were in both of the segmentation results obtained by the two image analysts (agreement) relative to the number of pixels that were not in both (disagreement). An overall agreement comparison was done by finding means and standard deviations of kappa statistics and the similarity measure found for each tissue type in the studies. The kappa statistics for white matter was the largest (0.80) followed by those of gray matter (0.68), partial volume (0.67), total lesion (0.66), and CSF (0.44). The similarity measure showed the same trend but it was always higher than kappa statistics. It was 0.85 for white matter, 0.77 for gray matter, 0.73 for partial volume, 0.72 for total lesion, and 0.47 for CSF.

A system that automatically segments and labels complete glioblastoma-multiform tumor volumes in magnetic resonance images of the human brain is presented. The magnetic resonance images consist of three feature images (T1- weighted, proton density, T2-weighted) and are processed by a system which integrates knowledge-based techniques with multispectral analysis and is independent of a particular magnetic resonance scanning protocol. Initial segmentation is performed by an unsupervised clustering algorithm. The segmented image, along with cluster centers for each class are provided to a rule-based expert system which extracts the intra-cranial region. Multispectral histogram analysis separates suspected tumor from the rest of the intra-cranial region, with region analysis used in performing the final tumor labeling. This system has been trained on eleven volume data sets and tested on twenty-two unseen volume data sets acquired from a single magnetic resonance imaging system. The knowledge-based tumor segmentation was compared with radiologist-verified `ground truth' tumor volumes and results generated by a supervised fuzzy clustering algorithm. The results of this system generally correspond well to ground truth, both on a per slice basis and more importantly in tracking total tumor volume during treatment over time.

The study presented in this paper tests the hypothesis that the combination of a global similarity transformation and local free form deformations can be used for the accurate segmentation of internal structures in MR images of the brain. To quantitatively evaluate our approach, the entire brain, the cerebellum and the head of the caudate have been segmented manually on one of the volumes and mapped back onto all the other volumes using the computed transformations. The contours so obtained have been compared to contours drawn manually around the structures of interest in each individual brain. Manual delineation was repeated to test intra-rater variability. Contours were quantitatively compared using a similarity index defined as 2 times the area encircled by both contours divided by the sum of the areas encircled by each contour. This index ranges from 0 to 1 with zero indicating zero overlap and one indicating a perfect agreement between two contours. It is sensitive to both displacement and differences in shape and it is thus preferable to a simple area comparison. Results indicate that the method we proposed can be used to segment accurately and fully automatically large and small structures in high resolution 3D images of the brain. The average similarity indices between the manual and automatic segmentations are 0.96, 0.97, and 0.845 for the whole head, the cerebellum, and the head of the caudate respectively. These numbers are 0.97, 0.97, and 0.88 when two manual delineations are compared. Statistical analysis reveals that the differences in mean similarity indices between the two manual delineations and between the manual delineations and the automatic segmentation method are statistically significant for the whole head and the caudate but not for the cerebellum. It is shown, however, that similarity indices in the range of 0.85 correspond to contours that are virtually undistinguishable.

A knowledge-based system has been developed for segmenting computed tomography (CT) images. Its modular architecture includes an anatomical model, image processing engine, inference engine and blackboard. The model contains a priori knowledge of size, shape, X-ray attenuation and relative position of anatomical structures. This knowledge is used to constrain low-level segmentation routines. Model-derived constraints and segmented image objects are both transformed into a common feature space and posted on the blackboard. The inference engine then matches image to model objects, based on the constraints. The transformation to feature space allows the knowledge and image data representations to be independent. Thus a high-level model can be used, with data being stored in a frame-based semantic network. This modularity and explicit representation of knowledge allows for straightforward system extension. We initially demonstrate an application to lung segmentation in thoracic CT, with subsequent extension of the knowledge-base to include tumors within the lung fields. The anatomical model was later augmented to include basic brain anatomy including the skull and blood vessels, to allow automatic segmentation of vascular structures in CT angiograms for 3D rendering and visualization.

To segment complex and versatile image data from different modalities it is almost impossible to achieve satisfying results without the consideration of contextual information. In this approach, image segmentation is regarded as a high- dimensional optimization task, that can be solved by stochastical methods like evolutionary algorithms (EA). Initially, the iterative algorithm is provided with a set of good-quality sample segmentations. An efficient EA-based learning strategy generates a segmentation for a given target image from the provided samples. This two-level process consists of a global image-based optimization whose convergence is enhanced by locally operating pixel-based Boltzmann processes which restrict the search space to reasonable subsets. The stochastic reconstruction extracts the relevant information from the samples in order to adapt it onto the current segmentation problem, which results in a consistent labeling for the target image. The algorithm works unsupervised, because the range of possible labels and their contextual interpretation is provided implicitly by the sample segmentations. To prove the usefulness of the method experimental results based on both, reproducible phantom images and physiological NMR scans are presented. Moreover, an analysis of the basic segmentation and convergence properties is provided.

For women undergoing assisted reproductive therapy, ovarian ultrasound has become an invaluable tool for monitoring the growth and assessing the physiological status of individual follicles. Measurements of the size and shape of follicles are the primary means of evaluation by physicians. Currently, follicle wall segmentation is achieved by manual tracing which is time consuming and susceptible to inter- operator variation. We are introducing a completely automated method of follicle wall isolation which provides faster, more consistent analysis. Our automated method is a 4-step process which employs watershed segmentation and a knowledge-based graph search algorithm which utilizes a priori information about follicle structure for inner and outer wall detection. The automated technique was tested on 36 ultrasonographic images of woman's ovaries. Five images from this set were omitted due to poor image quality. Validation of the remaining 31 ultrasound images against manually traced borders has shown an average rms error of 0.61 +/- 0.40 mm for inner border and 0.61 +/- 0.31 mm for outer border detection. Quantitative comparison of the computer-defined borders and the user-defined borders advocates the accuracy of our automated method of follicle analysis.

Clinicians have long desired early detection of neurological abnormality for treatment of brain malignancies. In attempts to address this concern, there are numerous reports publishing normative databases of age-related changes of the brain in healthy controls, many using magnetic resonance imaging (MRI). However, most of the method used to access tissue volumes were subject to observer variability. We developed a Kohonen self-organizing map to automatically segment MR images for reproducible and accurate identification of tissues. The developed method was applied to quantitatively assess subtle volume differences in normal controls due to maturational and degenerative changes. The volumes calculated in the test population of 73 controls agreed with current hypothesizes concerning age-related changes of the brain as determined by linear regression analysis of segmented tissue to age. Percent gray matter and percent white matter, as well as the ratio of gray matter to white matter were all found to be significantly correlated with age. Percent gray matter and the ratio of gray matter to white matter were inversely proportional to age while percent white matter was directly proportional to age. These results suggest the utility of the developed segmentation technique, as well as the clinical application it may hold.

Tumor segmentation from magnetic resonance (MR) images aids in tumor treatment by tracking the progress of tumor growth and/or shrinkage. In this paper we present an automatic segmentation method which separates non-enhancing brain tumors from healthy tissues in MR images. The MR feature images used for the segmentation consist of three weighted images (T1, T2 and proton density) for each axial slice through the head. An initial segmentation is computed using an unsupervised clustering algorithm. Then, integrated domain knowledge and image processing techniques contribute to the final tumor segmentation. The system was trained on two patient volumes and preliminary testing has shown successful tumor segmentations on four patient volumes.

Over the past decade, there has been increased interest in quantifying cell populations in tissue sections. Image analysis is now being used for analysis in limited pathological applications, such as PAP smear evaluation, with the dual aim of increasing for accuracy of diagnosis and reducing the review time. These applications primarily used gray scale images and dealt with cytological smears in which cells were well separated. Quantification of routinely stained tissue represented a more difficult problem in that objects could not be separated in gray scale as part of the background could also have the same intensity as the objects of interest. Many of the existing semiautomatic algorithms were specific to a particular application and were computationally expensive. Hence, this paper investigates the general adaptive automated color segmentation approaches, which alleviate these problems. In particular, competitive learning and the fuzzy-kohonen networks are studied. Four adaptive segmentation algorithms are compared using synthetic images and clinical microscopy slide images. Both qualitative and quantitative performance comparisons are performed with the clinical images. A method for finding the optimal number of clusters in the image is also validated. Finally the merits and feasibility of including contextual information in the segmentation are discussed along with future directions.

The determination of objective numerical criteria from nuclear medicine image data renders it possible to plan and control therapies, to compare inter- and intra-individual studies as well as time course studies and to facilitate the dominating visual interpretation of scintigrams. SPECT performs real 3D functional imaging of radionuclide distributions. The basic numerical value is the functional volume of a certain region. The volume is one prerequisite for activity measurement, but the value itself is of diagnostic importance, too. For determining the region boundaries several segmentation approaches are commonly used which are generally based on interactive ROI drawing, thresholding or edge detection methods. The image quality properties of SPECT render the segmentation process more difficult in any case. We propose an alternative segmentation approach where the crisp decision `object: yes or not' is substituted by a fuzzy boundary model `object: more or less'.

In this work, we describe an automated approach to morphometry based on spatial normalizations of the data, and demonstrate its application to the analysis of gender differences in the human corpus callosum. The purpose is to describe a population by a reduced and representative set of variables, from which a prior model can be constructed. Our approach is rooted in the assumption that individual anatomies can be considered as quantitative variations on a common underlying qualitative plane. We can therefore imagine that a given individual's anatomy is a warped version of some referential anatomy, also known as an atlas. The spatial warps which transform a labeled atlas into anatomic alignment with a population yield immediate knowledge about organ size and shape in the group. Furthermore, variation within the set of spatial warps is directly related to the anatomic variation among the subjects. Specifically, the shape statistics--mean and variance of the mappings--for the population can be calculated in a special basis, and an eigendecomposition of the variance performed to identify the most significant modes of shape variation. The results obtained with the corpus callosum study confirm the existence of substantial anatomical differences between males and females, as reported in previous experimental work.

Studies of shape, degree of folding and variability of the surface of the brain can be found as early as 1967. Measuring shape and degree of folding from radiological images is, however, a relatively recent development. Our objective is to develop a framework for the study of the shape and geometric properties of the brain surface from radiological images for the purposes of studying abnormalities of cortical folding. The construction of an explicit surface model allows the calculation of local measures of intrinsic curvature and folding, and provides a means of studying connectivity and shape variation. To speed up these calculations, we have developed a surface simplification method for discrete data representing surfaces derived from MR images. This method facilitates the calculation of these measures of interest while preserving the essential features of the surface in high curvature regions.

This paper describes a segmentation technique based on a new simplified active contour scheme. The method is a compromise between dynamic programming and the original `snake' solution. It follows a two-step procedure: the first step moves the contour points toward image features using a new potential vector field; the second step smoothes out the contour using a numeric filter to impose continuity and rigidity constraints on the contour model. This technique allows a natural definition of the active contour parameters in term of local curvature and rigidity of the contour. Since it performed particularly well in very noise images, the procedure is demonstrated for segmenting bone structures of the human spine from sagittal radiographs.

An important procedure in pediatrics is to determine the skeletal maturity of a patient from radiographs of the hand. There is great interest in the automation of this tedious and time-consuming task. We present a new method for the segmentation of the bones of the hand, which allows the assessment of the skeletal maturity with an appropriate database of reference bones, similar to the atlas based methods. The proposed algorithm uses an extended active contour model for the segmentation of the hand bones, which incorporates a-priori knowledge of shape and topology of the bones in an additional energy term. This `scene knowledge' is integrated in a complex hierarchical image model, that is used for the image analysis task.

In order to evaluate the proper breast implant, the surgeon relies on a standard set of measurements manually taken on the subject. This approach does not allow to obtain an accurate reconstruction of the breast shape and asymmetries can easily arise after surgery. The purpose of this work is to present a method which can help the surgeon in the choice of the shape and dimensions of a prosthesis allowing for a perfect symmetry between the prosthesis and the controlateral breast and can be used as a 3D visual feedback in plastic surgery.

A new approach to computer supported recognition of melanoma and naevocytic naevi based on high resolution skin surface profiles is presented. Profiles are generated by sampling an area of 4 X 4 mm² at a resolution of 125 sample points per mm with a laser profilometer at a vertical resolution of 0.1 micrometers . With image analysis algorithms Haralick's texture parameters, Fourier features and features based on fractal analysis are extracted. In order to improve classification performance, a subsequent feature selection process is applied to determine the best possible subset of features. Genetic algorithms are optimized for the feature selection process, and results of different approaches are compared. As quality measure for feature subsets, the error rate of the nearest neighbor classifier estimated with the leaving-one-out method is used. In comparison to heuristic strategies and greedy algorithms, genetic algorithms show the best results for the feature selection problem. After feature selection, several architectures of feed forward neural networks with error back-propagation are evaluated. Classification performance of the neural classifier is optimized using different topologies, learning parameters and pruning algorithms. The best neural classifier achieved an error rate of 4.5% and was found after network pruning. The best result in all with an error rate of 2.3% was obtained with the nearest neighbor classifier.

When interpreting and analyzing magnetic resonance angiography images, the 3D overall tree structure and the thickness of the blood vessels are of interest. This shape information may be easier to obtain from the skeleton of the blood vessels. Skeletonization of digital volume objects denotes either reduction to a 2D structure consisting of 3D surfaces, and curves, or reduction to a 1D structure consisting of 3D curves only. Thin elongated objects, such as blood vessels, are well suited for reduction to curve skeletons. Our results indicate that the tree structure of the vascular system is well represented by the skeleton. Positions for possible artery stenoses may be identified by locating local minima in curve skeletons, where the skeletal voxels are labeled with the distance to the original background.

We present an efficient algorithm for calculating the central path of a computer-generated colon model created from helical computed tomography image data. The central path is an essential aid for navigating through complex anatomy such as the colon. Our algorithm involves three steps. In the first step, we generate a 3D skeleton of the binary colon volume using a fast topological thinning algorithm. In the second step, we employ a graph search algorithm to remove extra loops and branches. These loops and branches are caused by holes in the object which are artifacts produced during image segmentation. In the final step, we compute a smooth representation of the central path by approximating the skeleton with cubic B-splines. This final step is necessary because the skeleton contains many abrupt changes in direction due to the discrete nature of image data. The user supplies two endpoints for the central path; otherwise, the algorithm is fully automated. Experimental results demonstrate that the algorithm is not only efficient but also robust. Use of this method in virtual endoscopy systems should have widespread clinical implications.

A 3D multiresolution segmentation approach based on a hierarchical watershed transformation (WST) on graphs for Computer Tomography images is presented. This approach is a 3D extension of the 2D WST on graphs which has already been successfully tested for the segmentation of CT images of the pelvis in hyperthermia planning. Analogous to the 2D technique the oversegmentation of the 3D WST is iteratively reduced through the application of the WST on graphs. Whereas for the 2D technique a stack of segmentation slices is constructed, here a stack of segmentation volumes consisting of 3D regions that correspond to 3D anatomical objects of different resolution in each volume are obtained. In contrast to the 2D approach which is at first applied on the image plane and hereinafter iteratively on graphs, the principle WST technique for graphs is now directly applied on the image plane. Besides the usage of a gradient with subpixel accuracy additionally a new procedure for the WST on this gradient is used. The labelled subpixel gradients are then used to label the voxels representing the segmentation result.

Determination of the surface of the heart is a challenging problem due to the heart's motion and the potential importance of subtle geometric features. The Meyer watershed has been shown to be an effective solution to this problem in images of the left ventricle in at least two medical image types. In this paper the technique is extended, first of all, by application to the right ventricle (RV) in an image from the Dynamic Spatial Reconstructor (DSR). Additionally, several important issues related to its general application are addressed including: (1)image anisotropy, (2)4D processing and (3) valve localization. The Meyer watershed model itself is also discussed in some detail with respect to its implementation and general properties. A simple by-product of surface reconstruction of the RV in the DSR images is that of the pulmonary arteries.

Early detection of colorectal polyps can improve morbidity and mortality due to cancer of the colon. The colon centerline can be used to expedite examination of the endoluminal surface for colorectal polyps. An automated technique has been developed that calculates the colon centerline from rectum to cecum from helical computed tomography slices of fully insufflated colons. Volume growing is initiated by indicating a seed point in the rectum, air voxels are grown and tagged with growth step numbers. The centers of mass of grown voxels with similar growth step numbers are used as a `forward' centerline. This procedure is repeated by growing from the cecum to the rectum to generate a `backward centerline'. The forward and backward centerlines are averaged to produce the calculated centerline. The technique was evaluated on a clinical colon case by comparing the calculated centerline with points indicated by 2 radiologists. Root mean square differences between the computed and indicated points were small (4 - 5 mm) and comparable to inter-observer differences. Results indicate that with this technique the centerline of the colon can be accurately and quickly calculated.

We present a surgical guidance system that incorporates preoperative image information (e.g. MRI or CT) and intraoperative ultrasound imaging to detect brain tissue deformation during image guided neurosurgery. Many interactive IGNS implementations involve using pre-operative image information (e.g. MRI or CT) as a guide to the surgeons throughout the procedure. Tissue movement during a procedure can be a significant source of error in these systems. By incorporating intraoperative imaging, the target volume can be scanned at any time, and mapped into the pre- operative image space. The surgeon can use this information to assess the accuracy of the guidance system at any time during the procedure. In addition, the system can be used to provide updated information of the progress of this procedure (e.g. extent of lesion removal). Validation results using a deformable multimodality imaging phantom are presented as well as initial examples of the system used in surgery.

To support ophthalmologists in their daily routine and enable the quantitative assessment of vascular changes in color fundus photographs vessel extraction techniques have been improved. Within a model based approach steerable filters have been tested for efficient and precise segmentation of the vessel tree. The global model comprises the detection of the optic disc, the finding of starting points close to the optic disc, the tracking of the vessel course, the extraction of the vessel contour and the identification of branching points. This helps evaluating image quality and pathological changes to the retina and thus, improves diagnosis and therapy. The optic disc location is estimated and then more precisely extracted with the help of a hierarchical filter scheme based on first- order gaussian kernels at varying orientations. Vessel points are automatically identified around the optic disc and the vessel course is tracked in the actual direction by second-order gaussian kernels at varying orientations and scales. Using this backbone, differently oriented first- order gaussian kernels approximate the vessel contour. Thus, the direction and diameter of each vessel segment are determined. Steerable filters enable most efficient implementation. The developed methods have been applied to color fundus photographs showing different levels of diabetic retinopathy.

Blood vessels overlying one another at distinct depths (and hence appearing to intersect) in the sclera of the eye can be distinguished reliably from those that in fact do branch within the same depth, using only the information contained in a single photograph of the conjunctiva. That conclusion arises from extension of earlier work that qualitatively inferred relative depth of vessels. The current research was motivated by the need to quantify such inferences in terms of their sensitivities and robustness. A physics first principles model forms the basis for selection of features that capture blood vessel depth information. Features extracted from the image are shown to be useful in that effort; their utility is verified with phantoms that mimic the behavior of the conjunctiva and sclera. Because no special preparations are needed, the method works as well on archived images as on newly-acquired ones, and thus can be used in retrospective studies of images of the eye and other diffuse media.

We developed a new method to compensate the rib structures in digital x-ray images. The intrinsic information of rib structures is eliminated and a higher image quality for the diagnosis of pulmonal structures is achieved. An essential task of the algorithm is the robust detection of the rib borders. In this paper we introduce three algorithms to perform this task. The first, introduced by Schreckenberg and Joswig, uses the hough transform to find rib borders, the second one uses a synergetic classifier to estimate the matching between rib edge templates and rib borders. The last one, the sinking lead algorithm, gives the best classification results by performing a matched template technique in combination with partial methods from the former two algorithms.

Magnetic resonance tagging has been shown to be a useful technique for non-invasively measuring the deformation of an in vivo heart. Tagged images appear with a spatially encoded pattern of lines called tag lines that move with the tissue and can be analyzed to reconstruct a description of left ventricular (LV) displacement during a portion of the cardiac cycle. Existing analysis methods require user- defined epicardial and endocardial contours. In this paper we present a method based on edge-preserving regularization techniques for reconstructing dense 2D left ventricular displacement field from tag line position data without prior knowledge of the LV contours. Our methods are demonstrated on both simulated and in vivo data.

The digital subtraction angiography (DSA) method is not directly applicable to cardiovascular sequences because of the anatomy motion. We have developed a simple cardiovascular extension of DSA that updates the mask in real time rather than applying the same mask on the whole sequence. The algorithm uses a velocity based segmentation to discriminate vessels and background. This discrimination is possible because of the cardiovascular sequences fundamental property of faster vessel motion compared to the background motion. The real time mask estimation is done as a recursively implemented generalized maximum operation over the sequence. This operation yields a sequence of masks which is subtracted from the original sequence. The algorithm is causal and can therefore be implemented in real time acquisition systems. We have applied it to x-ray fluoroscopic and radiographic cardiovascular sequences obtaining a nearly DSA-quality sequences with substantially improved vessel contrast. The algorithm de facto provides a simple cardia specific motion detection method which can be used in noise reduction algorithms. The recursive background estimation approach can be generalized to other cardiac imaging modalities.

By exploiting the correlation of ultrasound speckle patterns that result from scattering by underlying tissue elements, 2D tissue motion can be theoretically recovered by tracking the apparent movement of the speckle patterns. Speckle tracking, however, is an ill-posed inverse problem because of temporal decorrelation of the speckle patterns and the inherent low signal-to-noise ratio of medical ultrasonic images. This paper investigates the use of an adaptive deformable mesh for non-rigid tissue motion recovery from ultrasound images. The nodes connecting the mesh elements are allocated adaptively to stable speckle patterns that are less susceptible to temporal decorrelation. We use the approach of finite element analysis in manipulating the irregular mesh elements. A novel deformable block matching algorithm, making use of a Lagrange element for higher-order description of local motions, is proposed to estimate a non- rigid motion vector at each node. In order to ensure that the motion estimates are admissible to a physically plausible solution, the nodal displacements are regularized by minimizing the strain energy of the mesh deformations. Experiments based on ultrasound images of muscle contraction and on computer simulations have shown that the proposed algorithm can successfully track non-rigid displacement fields.

This paper describes a video analysis system, free of markers and set-up procedures, that quantitatively identified gait abnormalities in real-time from standard video images. A novel color 3D body model was sized and texture mapped to the exact characteristics of a person from video images. The kinematics of the body model was represented by a transformation tree to track the position and orientation of a person relative to the camera. Joint angles were used to track the location and orientation of each body part, with the range of joint angles being constrained by associating degrees of freedom with each joint. To stabilize tracking, the joint angles were estimated for the next frame. The calculation of joint angles, for the next frame, was cast as an estimation problem which was solved using an iterated extended Kalman filter. Patients with dopa-responsive parkinsonism, and age matched normals, were video taped during several gait cycles with walking movements successfully tracked and classified. The results suggested that this approach has the potential to guide clinicians on the relative sensitivity of specific postural/gait features in diagnosis.

A convenient and systematic protocol for evaluating the image quality of digitized mammographic phantom images has been developed. It involves the measurement of the contrast of a low-contrast nodule and/or a group of microcalcifications from images of the American College of Radiology mammographic accreditation phantoms acquired under different x-ray techniques.

Magnetic resonance imaging (MRI) is routinely used for the diagnosis of multiple sclerosis (MS), and for objective assessment of the extent of disease as a marker of treatment efficacy in MS clinical trials. The purpose of this study is to compare the evaluation of T2-weighted MRI scans in MS patients using a semi-automated quantitative technique with an independent assessment by a neurologist. Baseline, 6- month, and 12-month T2-weighted MRI scans from 41 chronic progressive MS patients were examined. The lesion volume ranged from 0.50 to 51.56 cm² (mean: 8.08 cm²). Reproducibility of the quantitative technique was assessed by the re-evaluation of a random subset of 20 scans, the coefficient of variation of the replicate determinations was 8.2%. The reproducibility of the neurologist evaluations was assessed by the re-evaluation of a random subset of 10 patients. The rank correlation between the results of the two methods was 0.097, which did not significantly differ from zero. Disease-related activity in T2-weighted MRI scans is a multi-dimensional construct, and is not adequately summarized solely by determination of lesion volume. In this setting, image analysis software should not only support storage and retrieval as sets of pixels, but should also support links to an anatomical dictionary.

In this paper, a local shape adaptive template filtering is proposed for an enhancement of signal-to-noise ratio in magnetic resonance imaging without degradation of resolution. Instead of conventional template filtering where template shape is fixed, multiple templates are defined in the proposed method and optimal template among the multiple templates is selected and applied in pixel-by-pixel basis. By the proposed scheme, edge blurring is minimized by selecting mostly matched template, yet SNR enhancement by the filter is fully utilized. Compared to existing 2D linear least square error filter or direction-adaptive recursive filter, the proposed shape-adaptive filter with multiple templates provides higher SNR and sharper edges for both natural images and artificial resolution phantom images.

Cellular Neural Networks (CNN) has become a very good mean for solution of the different kind of image processing problems. CNN based on multi-valued neurons (CNN-MVN) and CNN based on universal binary neurons (CNN-UBN) are the specific kinds of the CNN. MVN and UBN are neurons with complex-valued weights, and complex internal arithmetic. Their main feature is possibility of implementation of the arbitrary mapping between inputs and output described by the MVN, and arbitrary (not only threshold) Boolean function (UBN). Great advantage of the CNN is possibility of implementation of the any linear and many non-linear filters in spatial domain. Together with noise removing using CNN it is possible to implement filters, which can amplify high and medium frequencies. These filters are a very good mean for solution of the enhancement problem, and problem of details extraction against complex background. So, CNN make it possible to organize all the processing process from filtering until extraction of the important details. Organization of this process for medical image processing is considered in the paper. A major attention will be concentrated on the processing of the x-ray and ultrasound images corresponding to different oncology (or closed to oncology) pathologies. Additionally we will consider new structure of the neural network for solution of the problem of differential diagnostics of breast cancer.

New method of low-contrast radiological images analysis is presented. The key method's idea is based on representation of an observed image as the complex envelope of a hypothetical pseudocoherent wave field. The following analysis is based on the ideas of the virtual computer synthesis with using procedures for wave equation inversion and controlled interferometric synthesis. The method provides significant increase in the sensitivity and space resolving power of low-contrast images analysis.

We are developing a fully automated method for the detection of lung nodules in helical computed tomography (CT) images of the thorax. In our computerized method, gray-level thresholding is used to segment the lungs from the thorax region within each CT section. A rolling ball operation is employed to more accurately delineate the lung boundaries, thereby incorporating peripheral nodules within the segmented lung regions. A multiple gray-level thresholding scheme is then used to capture nodules by creating a series of binary images in which a pixel is turned `on' if the corresponding image pixel has a gray level greater than the selected threshold. Groups of contiguous `on' pixels are identified as individual signals. To distinguish nodules from vessels, geometric descriptors are calculated for each signal detected in the series of binary images. The values of these descriptors are input to an artificial neural network, which allows for the elimination of a high percentage of false-positive signals.

Neuroanatomical images previously used to study JPEG compression, were used to evaluate wavelet compression. Our purpose is twofold: (1) to evaluate wavelet compression on neurological images, (2) to compare the scores experts gave to the same images compressed by wavelets and JPEG techniques.

Previously, we have shown the effectiveness of using Bayesian image estimation (BIE) to reduce scatter and increase the contrast to noise ratio (CNR) in digital chest radiography without degradation of resolution. Here, we investigate the incorporation of a spatially varying scatter model. Previously, BIE used a simple model for scatter, where scatter was modeled as a spatially invariant radial exponential with a single full-width at half-maximum and magnitude. This invariance resulted in some overcompensation and some undercompensation. A new spatially varying scatter model, where each pixel can have a different scatter kernel magnitude, was incorporated into BIE and used to reduce scatter in quantitative chest radiographs. Scatter fractions were reduced to less than 3% in the lung and mediastinum at 8 iterations. The original BIE technique only reduced scatter fractions to less than 2% in the lung and 38% in the mediastinum. CNR was improved by approximately 60% in the lung region and 200% in the mediastinum. No degradation of resolution was measured. Visual inspection showed improvement of image quality. Incorporation of a spatially varying scatter model into BIE reduces scatter to levels which far exceed those provided by an anti-scatter grid and can increase CNR without loss of resolution.

A self-optimizing 3D deformable model has been developed which is able to segment branching anatomy. Its performance is compared with that of interactive segmentation in spiral CT of abdominal aortic aneurysms. SCT data from six individuals were selected retrospectively, representing a range of vascular geometry and tortuosity. As a reference, segmentation was performed twice by one observer using interactive 2D region growing. The self-optimizing 3D deformable model was applied twice, each with different initializations of the model in the aortic lumen. Dimensional and volume measurements were made, and boundary positions compared. The model was found to give qualitatively good representation, but was not able to follow vessels distal to the iliac bifurcation. The results agreed very well with the 2D interactive technique where structures ran orthogonal to the slice plane, with structures localized to 0.5 mm. The percentage difference in volume estimation between the model and the reference was 3% (the same as the agreement between the two reference segmentations). The mean closest distance between model and reference boundaries was 1.2 +/- 0.5 mm. Most discrepancies occurred at the bifurcations, and we conclude that the 3D deformable model requires further development for accurate representation of branching vascular structures in disease, but the accuracy of the model segmentation is sufficient for visualization or training.

Medical image segmentation has many applications, including tumor localization, radiation therapy planting, and 3D modeling, but its current use falls far short of its potential. To address this shortcoming, we are developing a unified software environment that facilitates the development and deployment of new and existing medical image segmentation algorithms, including classification-based, shape-based, region-based, edge-based, and hybrid algorithms.

Segmentation is a fundamental step in many quantitative analysis tasks for clinical ultrasound images. However, due to the speckle noises and the ill-defined edges of the object of interest, the classic image segmentation techniques are frequently ineffective in segmenting ultrasound images. It is either difficult to identify the actual edges or the derived boundaries are disconnected in the images. In this paper, we present a novel algorithm for segmentation of general ultrasound images, which is composed of two major techniques, namely the early vision model and the discrete snake model. By simulating human early vision, the early vision model can highlight the edges and, at the same time, suppress the speckle noises in an ultrasound image. The discrete snake model carries out energy minimization on the distance map rather than performing snake deformation on the original image as other snake models did. Moreover, instead of searching the next position for a snaxel along its searching path pixel by pixel, the discrete model only consider the local maxima as the searching space. The new segmentation algorithm has been verified on clinical ultrasound images and the derived boundaries of the object of interest are quite consistent with those specified by medical doctors.

The detection of left ventricular boundary is an interesting and challenging task in the cardiac analysis. In this paper, a self-learning contour finding model derived based on the snake model is designed to detect the echocardiac boundaries. The proposed model utilizes the genetic algorithms as a training kernel to acquire the weights for the driving forces in the snake deformation. Thus, the weights can be treated as a priori knowledge of contour definition before the contour finding process is proceeded. Both the synthetic and real image experiments are carried out to verify the performance of the proposed method.

Thresholding is one of the most common techniques for extracting objects from background because of its promising speed and simplicity. The images are usually ill-defined and the histograms are very noisy, therefore, the image thresholds are not easy to determine from the histogram directly. Based on fuzzy set theory, several fuzzy set algorithms have been reported to partition an image into objects and non-interesting regions to reduce the dimensionalities of the image. However, most of the existing approaches just choose the bandwidth of fuzzy membership functions experimentally/subjectively. In this paper, we propose a novel method to determine the bandwidth of fuzzy membership functions automatically. The optimal threshold is determined by choosing the proper bandwidth and minimizing the measure of fuzziness. The proposed approach has been tested on many images. The advantages of the proposed approach are: the bandwidth of fuzzy membership function is determined automatically and different bandwidths would be found between peaks so that the variations of the portions of the index-grey level graph can have different bandwidths. The images are well segmented, the details of the image are well preserved, and the segmented regions are homogeneous.

Image processing has to deal with many ambiguous situations. Fuzzy set theory is a good mathematical tool for handling the ambiguity or uncertainty. In order to apply the fuzzy theory, selecting the fuzzy region of membership function is an fundamental and important task. Most researchers use a predetermined window approach which has inherent problems. There are several formulas for computing the entropy of a fuzzy set. In order to overcome the weakness of the existing entropy formulas, this paper defines a new approach to fuzzy entropy and uses it to automatically select the fuzzy region of membership function so that an image is able to be transformed into fuzzy domain with maximum fuzzy entropy. The procedure for finding the optimal combination of a, b and c is implemented by a genetic algorithm. The proposed method selects the fuzzy region according to the nature of the input image, determines the fuzzy region of membership function automatically, and the post-processes are based on the fuzzy region and membership function. We have employed the novelly proposed approach to perform image enhancement and thresholding, and obtained satisfactory results.

In this paper, we present a method to classify hyperplastic and adenomatous polyps of large intestine semiautomatically. First, doctors locate the contour of the original polyp images by using other software package. We determine if there are gores on the polyp by using modified Sobel operator on eliminating specular reflection pixels of original color images. We then get the polyp's texture by summing the gradient magnitude of pixels within the polyps. After detecting the actual contour of the polyps, we can determined if the polyp's contour is obvious or not (i.e. if the polyp bulges smoothly or not). We then observe whether the polyp's color is redder than or whiter than its neighbors. Finally, we classify the polyp of the intestine by applying the above steps. The flow chart of classification is as shown. We apply our method on 77 color images with polyps of the intestine and compare the results with a doctor's diagnosis.

This paper describes the Digital Fundus Image Diagnostic System, a computerized image analysis system being developed to aid a human analyst, or grader, in the labeling and quantification of pathologic lesions indicative of background diabetic retinopathy. The grader uses a computer workstation to perform the analysis on a digitized color image of the ocular fundus (retina). The goal of the system is to provide precise quantitative measures of pathologic lesions (size, location, number and color), and place them in a computer database to be used for clinical records and for epidemiological studies.

Ultrasound may be used for imaging the trans-femoral residual limb in order to provide information for the improvement of prosthetic socket design. Compounding of several ultrasound B-scans is required for obtaining transverse images of the residual limb. In this paper, a method is presented by which a genetic algorithm is used to match B-scans taken in a horizontal plane around the residual limb for image compounding in order to reduce the effects of patient motion during scanning.

Rheumatoid arthritis of the hand can be characterized and assessed by the narrowing of the phalangeal joint spaces. These are ordinarily scored semi-quantitatively by a radiologist using radiographs of the hand. Software which delineates and measures the joint spaces would be a useful tool for diagnosis. The first part of such an algorithm has been developed which segments and identifies eight individual bones on digitized hand radiographs: the middle and proximal phalanges of the 2nd to 5th digits. The software also determines the locations of the distal interphalangeal, proximal interphalangeal, and metacarpophalangeal joint spaces for each digit.

We have developed a technique for fast and reliable, computer-assisted segmentation of the vessels, thereby obviating time-consuming manual segmentation of intracranial vessels for creation of a 3D model. The high quality of the bone segmentation greatly facilitates the segmentation of the vascular structures. As a result, computer tomography angiography examinations may be a viable alternative to a more invasive and expensive conventional angiography techniques used in the diagnosis of the pathology of intracranial vessels, especially in the cerebrovascular emergencies.

Adaptive histogram equalization (AHE) is a useful technique for expanding local contrast in medical images. The method is based on histogram equalization of each pixel based on a local NXN image region. If the number of pixels in the NXN local region is equal to the number of grayshades in the image, the equalized histogram can be directly constructed from the distribution histogram. This optimization may permit AHE to be performed on standard medical image display workstations.

Since its introduction to neurological surgery in the early 1980's, computer assisted surgery (CAS) with and without robotics navigation has been applied to several medical fields. The common issue all CAS systems is registration between two pre-operative 3D image modalities (for example, CT/MRI/PET et al) and the 3D image references of the patient in the operative room. In Wayne State University, a new way is introduced for medical image registration, which is different from traditional fiducial point registration and surface registration. We call it restricted surface matching (RSM). The method fast, convenient, accurate and robust. It combines the advantages from two registration methods mentioned before. Because of a penalty function introduced in its cost function, it is called `RSM'. The surface of a 3D image modality is pre-operatively extracted using segmentation techniques, and a distance map is created from such surface. The surface of another 3D reference is presented by a cloud of 3D points. At least three rough landmarks are used to restrict a registration not far away from global minimum. The local minimum issue is solved by use of a restriction for in the cost function and larger number of random starting points. The accuracy of matching is achieved by gradually releasing the restriction and limiting the influence of outliers. It only needs about half a minute to find the global minimum (for 256 X 256 X 56 images) in a SunSparc 10 station.

This paper presents a complete simulation of x-ray angiography, which we have used as tool to assist our research work in 3D reconstruction of vasculature from x-ray angiograms. X-ray angiography is an image acquisition technique routinely used in clinical practice to obtain images of blood vessel networks, which operates as follows: x-ray opaque dye is injected into the blood flow and is driven through the vessel network by time varying pressure gradients. An x-ray device captures a sequence of images as the dye flows, yielding an animation in which each image shows part of the vasculature. Once an animation from a given point of view has been acquired, the x-ray device is moved to a new position and the whole process is repeated. The final result is a set of animations. The simulation is able to reproduce all the salient points of x-ray angiography. Given our motivation (reconstructing vascular skeletons) we generalized the simulation so that it can produce static images of complete vasculature, and pre- segmented images that represented key stages in our reconstruction algorithm.

Characterization of pulmonary nodules plays a significant role in the differential diagnosis of lung cancer. This paper presents a method to characterize the internal structure of small pulmonary nodules through curvature-based descriptor using thin-section CT images. The present work is a first step toward the segmentation of the 3D nodule images by using a 3D deformable surfaces approach. Second, a curvature-based representation of the pulmonary nodule is derived. Based on this representation, the pulmonary nodules are globally characterized through the shape spectra. This quantification emphasizes the difference between benign and malignant pulmonary nodules surroundings. Experiments on true 3D nodule images demonstrate good performance of our curvature based analysis technique.

Medical images such as radiographs or micrographs provide valuable information about patient diseases. In spite of recent advances in imaging technologies and molecular probes, diagnosis is still performed by experts using visual interpretation. Quantitation of such visual tasks would render more objective data. For clinical application of image analysis, accurate determination of object boundary is often required and such a task is not trivial due to the complexity of biological objects. This paper presents a multi-resolutional watershed-based segmentation algorithm for expert users to extract object boundary from medical images reproducibly and accurately. Its accuracy is tested in medical images compared to the `marker-driven' gradient modification scheme.

In this study we compared trabecular bone mineral density (BMD) with textural parameters (cooccurence matrices features) extracted from trabecular bone structures in radiographic images of the hand. Our data consists of 12 cadaver hands radiographed and digitized. After application of a specific preprocessing step on all images, the textural parameters were calculated within 4 regions of interest defined within the metacarpal and proximal phalanges on trabecular bone. The results show that using a combination of textural parameters calculated at different directions within the ROI could increase significantly the correlation with BMD. Some further research will validate this finding on a larger set of data. This work is intended to be applicable in the study of bone fractures associated with osteoporosis, and could be of great benefit to a large segment of the population at risk.

This paper describes an intensity-based method for the segmentation of multiple sclerosis lesions in dual-echo PD and T2-weighted magnetic resonance brain images. The method consists of two stages: feature extraction and image analysis. For feature extraction, we use a ratio filter transformation on the proton density (PD) and spin-spin (T2) data sequences to extract the white matter, cerebrospinal fluid and the lesion features. The one and two dimensional histograms of the features are then analyzed to obtain different parameters, which provide the basis for subsequent image analysis operations to detect the multiple sclerosis lesions. In the image analysis stage, the PD images of the volume are first pre-processed to enhance the lesion tissue areas. White matter and cerebrospinal fluid masks are then generated and applied on the enhanced volume to remove non- lesion areas. Segmentation of lesions is performed in two steps: conspicuous lesions are extracted in the first step, followed by the extraction of the subtle lesions.

The speed and accuracy of Direct Fourier image reconstruction methods have long been hampered by the need to interpolate between the polar grid of Fourier data that is obtained from the measured projection data and the Cartesian grid of Fourier data that is needed to recover an image using the 2D FFT. Fast but crude interpolation schemes such as bilinear interpolation often lead to unacceptable image artifacts, while more sophisticated but computationally intense techniques such as circular sampling theorem (CST) interpolation negate the speed advantages afforded by the use of the 2D FFT. One technique that has been found to yield high-quality images without much computational penalty is a hybrid one in which zero-padding interpolation is first used to increase the density of samples on the polar grid after which bilinear interpolation onto the Cartesian grid is performed. In this work, we attempt to account for the success of this approach relative to the CST approach in three ways. First and more importantly, we establish that zero-padding interpolation of periodic functions that are sampled in accordance with the Nyquist criterion--precisely the sort of function encountered in the angular dimension of the polar grid--is exact and equivalent to circular sampling theorem interpolation. Second, we point out that both approaches make comparable approximations in interpolating in the radial direction. Finally, we indicate that the error introduced by the bilinear interpolation step in the zero- padding approach can be minimized by choosing sufficiently large zero-padding factors.

Tomosynthesis provides only incomplete 3D-data of the imaged object. Therefore it is important for reconstruction tasks to take all available information carefully into account. We are focusing on geometrical aspects of the scan process which can be incorporated into reconstruction algorithms by filtered backprojection methods. Our goal is a systematic approach to filter design. A unified theory of tomosynthesis is derived in the context of linear system theory, and a general four-step filter design concept is presented. Since the effects of filtering are understandable in this context, a methodical formulation of filter functions is possible in order to optimize image quality regarding the specific requirements of any application. By variation of filter parameters the slice thickness and the spatial resolution can easily be adjusted. The proposed general concept of filter design is exemplarily discussed for circular scanning but is valid for any specific scan geometry. The inherent limitations of tomosynthesis are pointed out and strategies for reducing the effects of incomplete sampling are developed. Results of a dental application show a striking improvement in image quality.

The study reported in this paper formulates image region detection problem in a multidimensional signal processing framework. Signal structure to sensor-array-processing presentation is created and the advanced sensor-array- processing technique is employed. Theoretical derivation of this method is given and simulation results are included. Results demonstrated: (1) this method can accurately detect the number of image regions, (2) it possesses extensive computer speed superiority over other existing methods, (3) it particularly facilitates fMRI activation foci detection.

The Gamma Knife (Elekta Instruments, Inc., Atlanta, GA), a neurosurgical, highly focused radiation delivery device, is used to eradicate deep-seated anomalous tissue within the human brain by delivering a lethal dose of radiation to target tissue. This dose is the accumulated result of delivering sequential `shots' of radiation to the target where each shot is approximately 3D Gaussian in shape. The size and intensity of each shot can be adjusted by varying the time of radiation exposure and by using one of four collimator sizes ranging from 4 - 18 mm. Current dose planning requires that the dose plan be developed manually to cover the target, and only the target, with a desired minimum radiation intensity using a minimum number of shots. This is a laborious and subjective process which typically leads to suboptimal conformal target coverage by the dose. We have used adaptive simulated annealing/quenching followed by Nelder-Mead simplex optimization to automate the selection and placement of Gaussian-based `shots' to form a simulated dose plane. In order to make the computation of the problem tractable, the algorithm, based upon contouring and polygon clipping, takes a 2 1/2-D approach to defining the cost function. Several experiments have been performed where the optimizers have been given the freedom to vary the number of shots and the weight, collimator size, and 3D location of each shot. To data best results have been obtained by forcing the optimizers to use a fixed number of unweighted shots with each optimizer set free to vary the 3D location and collimator size of each shot. Our preliminary results indicate that this technology will radically decrease planning time while significantly increasing accuracy of conformal target coverage and reproducibility over current manual methods.

The characterization of tumor aggressiveness is a very important step in cancer diagnosis and treatment. This can be achieved by examining the cells nuclei of the tissue. In order to perform a statistical study on a population of such nuclei, we have to characterize nuclei one by one. The problem is that the nuclei often appear in clusters and that other elements (totally unrepresentative of the studied tissue) can be present on the image. Moreover, we have to discard nuclei which do not guarantee valid data (broken nuclei, folded nuclei...). The purpose of our work is to separate clusters of nuclei into single-cell nuclei and eliminate undesirable elements of the image. Until now, this task was made by hand and was extremely slow and repetitive. Moreover, it implied a subjective basis, depending on the human operator. The method we developed is totally automatic. It is based on the elaboration of a binary mask containing objects which will be examined separately. Our algorithm has been tested on a large set of images coming from different tissues and the results are compared with the same task performed by human operators.

Multiresolution and multiorientation wavelet transforms (WTs), as the key CAD modules, were used in our previous study of CAD mass detection. The objective of this paper is to evaluate the roles of these WTs modules in the proposed CAD approach. A statistical analysis of the effects of WTs on image feature extraction for mass detection is taken including the effects of WTs on mass segmentation and a comparative study of discrimination ability of features extracted with WTs based and non-WTs based segmentation method. Three indexes are proposed to asses the segmentation. The effects of WTs on feature extraction are evaluated using ROC analysis of the feature discrimination ability. The statistical analysis demonstrates that the use of WTs modules results in a significant improvement in feature extraction for the previously proposed CAD mass detection method. The improvement, however, depends on the feature characteristics, large for boundary-related features while small for intensity-related features.

Diffusion process is an important tool for multiscale representation of images, image smoothing, enhancement and edge detection. In this paper, we propose a diffusion equation based method to enhance 3D CT images. The diffusion equation is applied in two ways. The first manner is to run the equation forward (diffusion) in the interior of regions, in hope that severe noise in these regions would be filtered out. The second way is to run the equation backward (antidiffusion) around edges to enhance contrast. The proposed algorithms are simple, and can be easily used in the practical applications. Finally, the algorithms were applied to real 3D CT images to demonstrate its effectiveness.

In this paper we address the problem of computing the point correspondences in a reference and subsequent image undergoing an orthogonal, affine, projective, or curved transformation with the aim to register the images. A whole image content based automatic algorithm for extracting point pairs from 2D monomodal medical images is presented. The selection of the most distinctive points from the reference image and the search for their pairs in the subsequent image have two things in common. First, the local operator by which the distinctive points are defined mimics the template matching used to find the point pairs. Second, the same similarity measure is used for both tasks. The results showed that the proposed automatic algorithm for point extraction is accurate and robust. Besides the efficient error handling, the algorithm has the capability to weight each pair according to the distinctiveness of the control point in the reference image, the similarity of the regions which surround the paired points, and the geometric consistency of the registered paired points.

In this paper, the Markov random field (MRF) clustering method for highly noisy medical image segmentation is presented. In MRF method, the image to be segmented is analyzed in a probabilistic way that establishes image model by a posteriori probability density function with Bayes' theorem, with relation between pixel positions as well as gray-levels involved. The adaptive threshold parameter is determined in the iterative clustering process to achieve global optimal segmentation. The presented method and other segmentation methods in use are tested on simulation images of different noise levels, and the numerical comparison result is presented. It also is applied on the highly noisy positron emission tomography images, in that the diagnostic hypoxia fraction is automatically calculated. The experimental results are acceptable, and show that the presented method is suitable and robust for noisy image segmentation.

In the clinical course of detecting masses, mammographers usually evaluate the surrounding background of a radiodense when breast cancer is suspected. In this study, we adapted this fundamental concept and computed features of the suspicious region in radial sections. These features were then arranged by circular convolution processes within a neural network, which led to an improvement in detecting mammographic masses.

This work focuses on the selection of a set of shape-based features in order to assist radiologists in differentiating between malignant and benignant clustered microcalcifications in mammograms. The results obtained allow the creation of a model for the evaluation of the benignant or malignant character of the microcalcifications in a mammogram, based exclusively on the following parameters: number of clusters, number of holes, area, Feret elongation, roughness and elongation. The performance of the classification scheme is close to the mean performance of three expert radiologists, which allows to consider the proposed method for assisting the diagnosis and encourages to continue the investigation in this field. Additionally, the work is based on an unpublished database formed by patients of the Regional Health Area of Girona, which in the future may contribute to increase the digital mammogram databases.

The purpose of this research is to characterize solitary pulmonary nodules as benign or malignant based on quantitative measures extracted from high resolution CT images. High resolution CT images of 17 patients with solitary pulmonary nodules and definitive diagnoses were obtained. The diagnoses of these 17 cases (11 benign and 6 malignant) were determined from either radiologic follow-up or pathological specimens. On the HRCT images, solitary nodules were identified using semiautomated contouring techniques. From the resulting contours, several quantitative measures are extracted related to the nodule's size, shape, density and texture. A stepwise discriminant analysis was performed to determine which combination of measures are best able to discriminate between the benign and malignant nodules. Using several selected features, a linear discriminant analysis was performed on the 17 cases. The preliminary discriminant analysis identified two different texture measures as the top features in discriminating between benign and malignant nodules. The linear discriminant analysis using these features correctly classified 16/17 cases (94.1%) of the training set. A less biased estimate, using jackknifed training and testing yielded 15/17 cases (88.2%) correctly classified. The preliminary results of this approach are very promising in characterizing solitary nodules using quantitative measures extracted from HRCT images.

Identification of anatomical structures in magnetic resonance (MR) images of the human brain is achieved either by manual delineation of by applying coordinate system transformations to map the brain to a pre-labeled atlas. Manual segmentation of 3D MR data is a tedious task made additionally difficult by limitations in visualization. Affine transforms, like the Talairach stereotaxic space, perform a linear scaling of the brain based on manually selected landmarks. This often results in unsatisfactory accuracy for structures further away from the selected landmarks, particularly in pathological cases. It is also based on the trivializing assumption that the brain can be represented as a linearly scalable structure. In the effort to achieve a more accurate and consistent labeling, an algorithm has been designed for the automated alignment of a pre-labeled 3D brain atlas with a sample MRI volume. Alignment is achieved by elastically warping a finite element model of the atlas. The deformation is driven by a set of displacement constraints on the surface of individual brain structures. Solving this model results in a 3D displacement field for the entire atlas brain that aligns the segmented brain structure while extrapolating the deformation field to neighboring structures. The use of finite element modeling assures that this extrapolation occurs in a physically meaningful manner. The algorithm's performance was tested by matching the atlas image to warped versions of itself and to an individual sample brain. The amount of structural overlap achieved by a linear Talairach transform is also given for comparison. Elastic warping showed better performance compared to an affine transform alone or the Talairach method. Overlap increases with subsequent iterations with improvement directly related to the amount of model deformation.

In clinical cytology quantitative parameters have to be extracted from a large number of biological samples to obtain diagnostically relevant and reproducible information. Computer-assisted microscopy can provide methods that increase the quality and comparability of clinical studies by reducing the subjective influence of human operators on their results. In order to guarantee the correctness of extracted parameters automatic and reliable segmentation of the samples is required. For the detection of cytological objects a novel deformable membrane model is presented which is strictly based on macroscopical mechanics and statics. This is appropriate for modeling physiological membranes, because their shape is determined exclusively by mechanical forces. The self-driven membrane converges iteratively towards a stable state, where the contrary forces are in balance. However, active contours may not yield sufficient detection quality for acquisition of quantitative parameters. Therefore, after convergence a stochastic optimization process corrects the contour according to local graylevel information. This yields a contour that is well- adapted to the local graylevel structure. Additionally, for subsequent cytometric quantifications a local measure of confidence is provided for the contour. this can be used to enhance the robustness of the extracted parameters by incorporating the confidence factors in the quantification process. The method is applied to cytological and histological samples of different magnification.

An important goal to achieve in order to improve the clinical usefulness of regional cerebral blood flow (rCBF) SPECT studies is to register each brain to a standard anatomical atlas. For this purpose, we propose a simple three steps method. First the mid-sagittal plane is computed based on the left-right symmetry of the brain. Then the AC- PC line (main axis of the Talairach & Tournoux reference system) is obtained from the positions of four landmarks in the mid-sagittal plane. Finally, three linear scaling parameters are determined to adjust the size of the subject brain within the Talairach & Tournoux Atlas. The method was successfully validated with a set of 64 X 64 X 64 SPECT Monte-Carlo simulations of the brain.

Quantitative coronary angiography (QCA) diameter measurements have been used as an endpoint measurement in clinical studies involving therapies to reduce coronary atherosclerosis. The accuracy and precision of the QCA measure can affect the sample size and study conclusions of a clinical study. Measurements using x-ray test phantoms can underestimate the precision and accuracy of the actual arteries in clinical digital angiograms because they do not contain complex patient structures. Determining the clinical performance of QCA algorithms under clinical conditions is difficult because: (1) no gold standard test object exists in clinical images, (2) phantom images do not have any structured background noise. We purpose the use of computer simulated arteries as a replacement for traditional angiographic test phantoms to evaluate QCA algorithm performance.

We have been developing a computerized scheme to restore cone-beam CT images that are degraded by scatter radiation. In reconstructing 3D image, the cone-beam CT scanner has some advantages over helical CT scanner in shorter reconstruction time and in higher spatial resolution. However, because large amount of scatter radiation is included in 2D projection data, the quality of reconstructed image is worse and cupping artifacts appear in CT images. We have tried to improve the quality of reconstructed images using the multiresolution analysis. To achieve this, the projection data was decomposed into several signals with different spatial resolution and then reconstructed from these decomposed signals by multiplying weighting factors. Moreover, using these weighting factors, we have designed the optimal filter to reconstruct the CT images without any extra computing time. Our results showed that the CT images were enhanced the contrast and improved the quality.

Our project is concerned with the improvement of radiation treatment procedures for ocular tumors. In this context the application of proton beams offers new possibilities to considerably enhance precision and reliability of current radiation treatment systems. A precise model of the patient's eye and the tumor is essential for determining the necessary treatment plan. Current treatment systems base their irradiation plan calculations mainly on schematic eye models (e.g., Gullstrand's schematic eye). The adjustment of the model to the patient's anatomy is done by distorting the model according to information from ultrasound and/or CT images. In our project a precise model of the orbita is determined from CT, high resolution MRT, ultrasound (A-mode depth images and/or 2D B-mode images) and photographs of the fundus. The results from various segmentation and image analysis steps performed on all the data are combined to achieve an eye model of improved precision. By using a proton cannon for the therapy execution, the high precision of the model can be exploited, thus achieving a basic improvement of the therapy. Control over the destruction of the tumor can be increased by maximizing the dose distributions within the target volume keeping the damage in the surrounding tissue to a minimum. This article is concerned with the image processing to generate an eye model on which treatment planning is based.

The presented paper treats a subject of elaboration of new algorithms for recognition of lesions and analysis of shape features of selected abdominal cavity organs visible on radiograms or tomograms. The aim of the methods is to determine and examine morphological shapes of the analyzed anatomical structures in order to diagnose cancerous lesions and inflammatory processes. The formulated target was accomplished in the case of the diagnosis of cancer and chronic inflammation of the pancreas made on the base of X- ray images obtained during the ERCP examinations. For this purpose an effective algorithm for thresholding of the ERCP images was employed. Hence it was possible to extract the pancreas duct together with morphological changes which could occur. Then, thanks to determination and application of special sequence of geometric operations (skeletonizing and rotations of contour points about a skeleton), a linear graph representing the width of pancreas duct and showing morphological changes was obtained. In order to find these changes the context-free attributed grammars, enabling description of all searched morphological changes were used. These attributes contained an additional information (height and width of the discovered change) used for recognition of ambiguous cases. For proper description and recognition of symptoms, for which the 2D analysis is required (i.e. e.g. large cavernous bulges), the language of shape features description with a special multidirectional sinquad distribution were employed. Research on usefulness of the proposed methods, performed so far, justified the application of syntactic methods to recognition of medical images, especially to support medical diagnostics.

Analyzing medical images, which have been stored in digital information system day by day, is expected to make it possible to formulate knowledge useful for image diagnosis. In this paper, we propose a method for unsupervised medical image segmentation as the pre-processing of the analysis aiming to clear the relation between the image features and the possible outcome of a medical condition. In the proposed method, a square region around the every pixel is considered as a pattern vector, and a set of pattern vectors acquired from whole image are classified by using the technique of hierarchical clustering. In the hierarchical clustering, the set of pattern vectors is divided into two clusters at each node, according to the statistical criterion based on the entropy in thermodynamics. Results on the test image generated by the Markov Random Field model and the real medical images, photomicrographs of intestine, are shown.

In this paper, we report the image processing technique for computer-aided diagnosis of lung cancer screening system by CT (LSCT). LSCT is the newly developed mobile-type CT scanner for the mass screening of lung cancer by our project team. In this new LSCT system, one essential problem is the increase of image information to be diagnosed by a doctor to about 30 slices per patient from 1 X-ray film. To solve this difficult problem, we are trying to reduce the image information drastically to be displayed for the detector by image processing techniques. We propose a new method named Variable-New-Quoit filter for the automatic recognition of the pathological shadow candidates. Our computer aided diagnosis system can satisfactorily reduce the number of CT cross sections by this method, containing the abnormal shadow candidates.

The quality of Magnetic Resonance (MR) images depends on several factors: static and radiofrequency field nonuniformities, gradient field nonlinearities, random noise, etc. Distortion correction and noise filtering not only increase the image visual quality but are indispensable steps prior to further applications, such as: image segmentation, quantitative analysis, 3D visualization, surgical planning, stereotactic neurosurgery. Imaging experiments were performed on a 0.1T MR scanner (Giroimag 02), developed in our center, using a head coil. Intensity and geometric distortions were corrected using correction maps obtained from images of suitable phantoms. The applied post-processing methods improved the image quality. Uniformity was increased significantly (about 20%) with the obtained correction matrix. Geometric correction reduced the distortion to sub-millimeter values. Significant improvements in S/N ratio were obtained by means of the noise filtering procedures tested. The edge information was preserved acceptably. Adaptive nonlinear filters (modified Gaussian filter) and Wavelet Denoising provided the best results increasing the image S/N ratio in more than 90% with relative minor resolution loss. All these techniques were implemented and tested upon a wide range of images and form a set of tools incorporated to the Image Acquisition System for Cuban MR scanners Giroimag.

Methods based on mutual information have shown promising results for matching of multimodal brain images. This paper discusses a multiscale approach to mutual information matching, aiming for an acceleration of the matching process while considering the accuracy and robustness of the method. Scaling of the images is done by equidistant sampling. Rigid matching of 3D magnetic resonance and computed tomography brain images is performed on datasets of varying resolution and quality. The experiments show that a multiscale approach to mutual information matching is an appropriate method for images of high resolution and quality. For such images an acceleration up to a factor of around 3 can be achieved. For images of poorer quality caution is advised with respect to the multiscale method, since the optimization method used (Powell) was shown to be highly sensitive to the local optima occurring in these cases. When incorrect intermediate results are avoided, an acceleration up to a factor of around 2 can be achieved for images of lower resolution.

We present a novel algorithm for obtaining fuzzy segmentations of images that are subject to multiplicative intensity inhomogeneities, such as magnetic resonance images. The algorithm is formulated by modifying the objective function in the fuzzy c-means algorithm to include a multiplier field, which allows the centroids for each class to vary across the image. First and second order regularization terms ensure that the multiplier field is both slowly varying and smooth. An iterative algorithm that minimizes the objective function is described, and its efficacy is demonstrated on several test images.

In this paper, we introduce a new segmentation technique (called Kohonen snake) based on the neural simulation of deformable models designed to reconstruct 3D objects. Kohonen snake possesses all properties of Kohonen networks (lateral interaction during the learning process, topologically preserving mapping) and of deformable models (namely, elastic properties). Elastic properties of the physics-based Kohonen ring improves the shortcomings of the Kohonen network related to twisting, `dead' neurons, accumulation and rounding the network, whereas the data- driven approach of Kohonen snake improves the problem of initialization and local minima of the snakes. When integrating both models, the first question is how to combine their parameters. We simulate the Kohonen snake behavior with different parameter values using sequential and parallel weight updating, study the need of decreasing the parameters and of reordering image features. As a result, we conclude that Kohonen snake has better control on its shape that makes it less dependent on the values of its parameters and initial conditions. Our tests on segmentation of synthetic and real images illustrate the usefulness of the Kohonen snake technique.

An adaptive coding is proposed for magnetic resonance (MR) imaging. In this paper, we propose an adaptive coding using the matching pursuit algorithm to the MR imaging. The matching pursuit algorithm is performed with the dictionary consisting windowed Fourier basis set to achieve adaptive coding. Image domain is segmented into several groups. Since the image signal is divided into small scale, the matching pursuit with the segmented signal is more effective than that with whole image domain. Since matching pursuit is a greedy algorithm to find waveforms which are the best match for an object-signal, the signal can be decomposed with a few iterations thereby resulting reduction of imaging time. Computer simulations and experiments are performed to verify the proposed technique.

In this paper, adaptive lossy coding procedures are applied to 12 bit/pixel lung radiograms, with the general goal to achieve high compression ratios while preserving the diagnostic information. To grant the diagnostic accuracy, regions that are classified as significant from a pathological point of view, have been preserved in the coding process. The extraction of relevant regions (nodules) has been performed automatically by the use of suitable operators based on mathematical morphology. Image structures, are classified in different categories, to treat the `generic' ones, peculiarly bony structures prevailing in this kind of images, differently from those having a diagnostic interest, because joined to specific pathologies. The last version of the JPEG sequential coding scheme, that allows variable quantization, has been used to code the images. This procedure is based on the DCT transform applied to 8 X 8 image blocks followed by adaptive quantization of the transformed blocks and then by an entropy coding of the quantized coefficients. As the information loss is due to the quantization operation, blocks containing `interesting' structures are quantized using smaller steps.

A recent trend is the automatic screening of color ocular fundus images. The examination of such images is used in the early detection of several adult diseases such as hypertension and diabetes. Since this type of examination is easier than CT, costs less, and has no harmful side effects, it will become a routine medical examination. Normal ocular fundus images are found in more than 90% of all people. To deal with the increasing number of such images, this paper proposes a new approach to process them automatically and accurately. Our approach, based on individual comparison, identifies changes in sequential images: a previously diagnosed normal reference image is compared to a non- diagnosed image.

Virtual bronchoscopy is one method using a virtual reality equipment to improve diagnostical information drawn from CT data. As a preprocessing step the bronchial tract has to be extracted from the data. Therefore a method has been developed, which allows automatic segmentation of the bronchial tract. In the preprocessing step a backbone to the bronchial tract is extracted, which allows using more sophisticated methods for segmentation. In the second step either minimal variance region growing or an Active Contour Model is used. The first one uses statistical information to extend the presegmented area according to the local greyvalue distribution. The second method automatically fits the predefined contour to the local gradient information.

Quantitative vascular analysis is useful for treatment planning and evaluation of atherosclerosis, but it requires accurate and reliable determination of the 3D vascular structures from biplane images. To facilitate vascular analysis, we have developed technique for reliable estimation of the biplane imaging geometry as well as 3D vascular structures without using a calibration phantom. The centerlines of the vessels were tracked, and bifurcation points and their hierarchy were then determined automatically. The corresponding bifurcation points in biplane images were used to obtain an estimate of the imaging geometry with the enhanced Metz-Fencil technique, starting with an initial estimate based on gantry information. This initial estimate was iteratively refined by means of non-linear optimization techniques that aligned the projections of the reconstructed 3D bifurcation points with their respective image points. Methods have also been developed for assessing the accuracy and reliability of the calculated 3D vascular centerlines. Accuracy was evaluated by comparison of distances within a physical phantom with those in the reconstructed phantom. The reliability of the calculated geometries and 3D positions were evaluated using data from multiple projections and observers.

In this paper, we present a lapse observation algorithm for lung cancer detection using 3D thoracic helical CT images. Purpose of this study is to detect the interval change that exist between time different images of same patient. We employed two methods to detect the interval change. The first method is 3D local template matching method. We tried to detect the movement of blood vessel and other organs by this method. The second method is subtraction method. We tried to detect the new shadow by this method. The subtraction technique is the common method which detects the interval change. If the two images are produced in an identical manner, the subtraction image derived from a pair of thin slice CT with time difference having an uniform zero pixel value except for regions with interval changes. However, the same 3D thoracic images are not obtained. Therefore, we study the method to correct the location between 3D thoracic images with time difference.

This paper presents development and application of an automated scheme to minimize user dependency of the eigenimage filter. The steps of the new method are as follows: (1) User defines sample regions of interest on central location of the volume and generates the corresponding eigenimages. (2) Original images are segmented using a self organizing data analysis technique. (3) Regions for the tissue types are automatically found from the segmentation results. (4) Signature vectors are estimated from these regions and are compared with those obtained from the user initialization. If they are similar, these signature vectors are used to run eigenimage filtering. If they are not similar, the clustering parameters are adjusted and the procedure is repeated until similar signatures are found. (5) Next slice is loaded and signature vectors from the previous slice are used to get initial eigenimages. Cluster analysis is used to generate regions. The procedure described in the previous step is repeated until similar signatures are found. Then, final eigenimages are obtained. (6) Previous step is repeated until the last slice of the volume is analyzed. (7) Volume of each tissue type is estimated from the resulting eigenimages. Details and significance of each step are explained. Experimental results using simulation, phantom, and brain images are presented.

This paper addresses the problem of fusing the information content of two uncalibrated sensors. This problem arises in registering images of a scene when it is viewed via two different sensory systems, or detecting change in a scene when it is viewed at two different time points by a sensory system (or via two different sensory systems or observation channels). We are concerned with sensory systems which have not only a relative shift, scaling and rotational calibration error, but also an unknown point spread function (that is time-varying for a single sensor, or different for two sensors). By modeling one image in terms of an unknown linear combination of the other image, its powers and their spatially-transformed (shift, rotation and scaling) versions, a signal subspace processing is developed for fusing uncalibrated sensors. Numerical results with realistic 3D magnetic resonance images of a patient with multiple sclerosis, which are acquired at two different time points, are provided.

In this paper we explore the use of feature selection techniques to improve the generalization performance of pattern recognizers for computer-aided diagnosis. We apply a modified version of the sequential forward floating selection (SFFS) of Pudil et al. to the problem of selecting an optimal feature subset for mass detection in digitized mammograms. The complete feature set consists of multi-scale tangential and radial gradients in the mammogram region of interest. We train a simple multi-layer perceptron (MLP) using the SFFS algorithm and compare its performance, using a jackknife procedure, to an MLP trained on the complete feature set (35 features). Results indicate that a variable number of features is chosen in each of the jackknife sets (12 +/- 4) and the test performance, A_z, using the chosen feature subset is no better than the performance using the entire feature set. These results may be attributed to the fact that the feature set is noisy and the data set used for training/testing is small. We next modify the feature selection technique by using the results of the jackknife to compute the frequency at which different features are selected. We construct a classifier by choosing the top N features, selected most frequently, which maximize performance on the training data. We find that by adding this `hand-tuning' component to the feature selection process, we can reduce the feature set from 35 to 8 features and at the same time have a statistically significant increase in generalization performance (p < 0.015).

In this paper, we focus on the 3D registration between the MR and SPECT brain images. It is undoubted that the human brain does not change its shape during imaging processes, registration of these two image sets can be modeled as a rigid body motion problem. Since the size of each voxel in MR and SPECT images is given, our goal is to find out three translation parameters and three rotation parameters of the motion and then match the object to a fixed position. In this paper, a hybrid method composed of homologous feature registration and surface fitting is developed. The centers of the two eyeballs are used as the feature sets for registration. Therefore, we can solve the three translation parameters and two rotation parameters. We then reslice the volume data. The remaining rotation parameter is determined by minimizing the surface distance, in the resliced 2D images. After obtaining all these parameters, the transformation matrix is determined, and the registration is well done.

We devised, built and tested a detector and segmentor of microcalcifications in mammograms. Our segmentor includes preprocessing, extraction of 17 features, genetic solution of the best subset of six features, and a k-nearest neighbor classifier to suppress false candidates.

To diagnose the lung cancer as to determine if it has malignant or benign nature, it is important to understand the spatial relationship among the abnormal nodule and other pulmonary organs. But the lung field has very complicated structure, so it is difficult to understand the connectivity of the pulmonary organs and the abnormal nodule for even medical doctors. In this paper, we describe a 3D image analysis method of the pulmonary organs using thin-section CT images. This method consists of two parts. The first is the classification of the pulmonary structure based on the anatomical information. The second is the quantitative analysis which is then applicable to differential diagnosis, such as differentiation of malignant or benign abnormal tissue.