1.

Introduction

The last several years have witnessed the development of numerous methods and techniques for obtaining two principal optical properties of tissue: the absorption coefficient (${\mu }_{\mathrm{a}}$) and transport (reduced) elastic scattering coefficient (${\mu }_{\mathrm{s}}^{\prime }$).^1–3 These parameters are used in the clinical context to: (1) diagnose pathological conditions or metabolic states and (2) monitor efficacy of therapeutic treatments. Optical absorption reflects the process of light attenuation by physiologically relevant chromophores, such as hemoglobin, melanin, carotenes, lipids, and water. Using a wavelength-dependent linear-square fit of absorption coefficient values and the Beer–Lambert law, the concentrations of individual chromophores can be calculated.^4–7 Additionally, optical scattering by tissue reflects the probability of light being deflected by its structural components, such as interstitial water, cell membranes, nuclei, and other organelles, as well as structural fibers.^8–11 Thus, macroscopic structural variation of tissue can be derived from scattering coefficients (amplitude and power) by Mie theory,¹² characterized by an exponential monotone decrease, relative to wavelength. Hence, the separate measurement of absorption and scattering parameters as a function of wavelength can provide a comprehensive biochemical, morphological, and histochemical characterization of the tissues under study. This broad range of tissue parameters interrogated may be used to noninvasively diagnose disease even in very early stages.

The range of optical methods applied to determine the optical properties of biological materials may be broadly divided into direct and indirect techniques. Direct methods rely directly on the experimental results, without any model of photon migration. Indirect methods are more complicated in their use, involve mathematical models and/or simulations of light propagation, and require sophisticated (and sometimes expensive) instrumentation.^13–16 Time-resolved diffuse reflectance,^17,18 frequency domain,^19–21 spatially resolved continues wave,^22–26 low coherence interferometry,^27,28 and oblique incidence reflectometry^29,30 are among the methods used to extract optical properties of sample tissue. Each of these methods has varying capabilities regarding accuracy, levels of noise and artifacts, field of view (FOV), spatial and temporal resolution, speed, penetration depth, etc. Among techniques recently developed for optical characterization of turbid media is spatially modulated illumination, known also as spatial frequency domain.^31,32 With this technique, periodic illumination (sinusoidal grid) patterns are projected to separately map tissue absorption and reduce light scattering over a wide FOV, in a noncontact and scan-free fashion. The advantages of this imaging technique include capability of depth sectioning, as well as wide-field and wild-range imaging, ease of use, and relatively low cost due to its minimal number of optical elements. The technical capabilities of spatially modulated illumination have been well documented over the last 12 years, with successful demonstration in phantom, animal, and human studies.^33–40

In the context of these contemporary techniques, this work aimed to apply structured illumination to acquire optical properties of tissue-simulating phantoms and human tissue in the near-infrared (NIR) region (600 to 1000 nm). As mentioned above, the application of spatially modulated illumination to quantify tissue properties was recently established in imaging setup and spectroscopy by several research groups. The innovation of the present study lies in the combination of both spectrometer and camera channels delivering simultaneous readings and in the ease of deriving optical properties. In this hybrid apparatus, the sinusoidal patterns are projected sequentially at low and high spatial frequencies (0 and $0.27{\mathrm{mm}}^{-1}$, respectively), and the diffuse reflected light is recorded simultaneously by a single noncontact spectrometer and two separate CCD cameras. The system’s distinct advantage over other methods is that it allows each method of signal capture to complement gaps in the other’s capabilities, delivering clinically relevant information regarding tissue properties at high spatial and spectral resolution. This study builds on previous work advancing optical spectroscopy of tissue by sinusoidal patterns of spatially modulated light³¹ and, to the best of our knowledge, is the first to combine such an illumination setup with parallel recording by complementary spectrometer and camera array channels.

The outline of the present paper is as follows: in Sec. 2, we provide details of the system operation, tissue property quantification analysis, and the calibration procedure to account for the system response. Section 3 will be devoted to experimental results measured from both phantoms and the human hand, as well as the calculation of tissue hemoglobin oxygen saturation. Additionally, we discuss the feasibility of reconstructing optical properties with recent analytical camera modeling.⁴¹ Finally, in Sec. 4, we discuss the results, future development, and the potential clinical and research applications of the technique.

2.

Experimental Study

2.2.

System Description

The experimental system is drawn schematically in Fig. 1(a). The system integrates two channels: spectrometer and CCDs fed by the same sinusoidal fringe illumination. Optical measurement is configured as follows: sinusoidal patterns (fringes), generated in MATLAB^® software, are serially projected onto the sample by a portable LED-based picoprojector (Aiptek, V10, LCoS Technology), passing through a collimated achromatic lens system (LS1). The collimated patterns are serially projected four times in total: three times at a fixed spatial frequency of $0.27{\mathrm{mm}}^{-1}$, each time at different phases (0 deg, 120 deg, and 240 deg) to determine the reduced scattering coefficient, followed by a final illumination at zero spatial frequency ($0{\mathrm{mm}}^{-1}$) without any phase shift (planar unmodulated illumination) for determination of the absorption coefficient. Demonstration of reflection patterns at these frequencies is shown in Fig. 1(b). While higher spatial frequency patterns are primarily sensitive to scattering changes but insensitive to absorption, the lower spatial frequency patterns are far more sensitive to changes in absorption than are they to scattering.^32,42,43 As illustrated, the diffusely reflected light is recorded simultaneously by both CCD cameras and the spectrometer, which view the same region through the beam splitters (BSs) in a noncontact fashion. In this way, the diffusely reflected light emanating from the sample is first split into two separate optical paths by BS1: the first propagate to a portable compact fiber-coupled spectrometer (USB4000-VIS-NIR, Ocean Optics) and the second propagate to BS2 where light is divided again before detection by two separate CCD cameras (Guppy PRO, F031B, Allied Vision Technology). Each camera, capable of imaging up to 123 frames per second at full resolution, produces a 14-bit monochrome image with $656(\mathrm{H})\times 492(\mathrm{V})$ pixel spatial resolution, at pixel size of $5.6\mu \mathrm{m}$. Each camera is equipped with a zoom lens system (Computar, 75 to 150 mm, f#/2.8, New York) and an optical bandpass filter (10-nm FWHM BPF, Thorlabs) of 650 nm (BP1) or 800 nm (BP2), respectively, placed in the front of the camera, enabling dual-band imaging. Thereby, the diffusely reflected light is first collected concomitantly by both channels and then filtered by each respective BPF prior to simultaneous image capture of the identical field. Camera exposure time and gain were manually adjusted for each wavelength filtering during the calibration process; both settings were found to be $\sim 18\%$ higher at 800 nm than at 650 nm because of the reduced quantum efficiency of the CCD at longer wavelengths. While originally obtained images were of high signal-to-noise ratio, collected diffuse images were nonetheless normalized prior to processing to decrease discrepancies in pixel intensity of the filtered images. The delivered light to the spectrometer from BS1 was filtered by adjusted pinholes and focused by a second lens system (LS2) onto a single optical fiber, ($D=400\mu \mathrm{m}$, $\mathrm{NA}=0.22$), coupled to a connection interface to the spectrometer, measuring at a spectral range of 400 to 1000 nm. The optical fiber, placed on the center of BS1 and LS2, is positioned by a metallic rod attached to an $x\text{-}y\text{-}z$ mechanical stage translation for position control to ensure maximum light collection. In contrast to the CCDs, the spectrometer recorded a range of wavelengths that dramatically increased the spectral resolution. However, the improved spectral resolution came at the expense of the field of vision imaged since single-point measurements cannot provide spatial information. The spectrometer interfaces to the computer via USB 2.0 port and is controlled by Ocean Optics software (OOI Base 32). Each set of data acquired by the spectrometer includes the measurements of replicate spectra ($n$) during a single data collection session. These spectra are later assembled together to a single average spectrum, thereby increasing the signal-to noise ratio by a factor of $\surd n$. The average time required to acquire one complete measurement was 10 s. The entire setup is controlled by MATLAB^® software (MathWorks, Massachusetts) and imaging acquisition, synchronization, and data processing are achieved using in-house developed MATLAB^® scripts on a laptop computer equipped with an Intel^® Core™ $i5$-3210M processor running at 2.5 GHz with 4 GB memory.

Fig. 1

(a) Sketch of the combined optical arrangement used in the current study. (b) Example of projected structured pattern at $f_x=0{\mathrm{mm}}^{-1}$ (DC frequency, left) and when $f_x=0.27{\mathrm{mm}}^{-1}$ at phase shift (AC frequency, right), as captured by one of the CCD camera.

2.3.

Data Analysis

Each sample was illuminated at low ($0{\mathrm{mm}}^{-1}$) and high ($0.27{\mathrm{mm}}^{-1}$) spatial frequencies to determine the absorption and reduced scattering coefficients, respectively. Consequentially, measurements taken at these two extreme spatial frequencies resulted in contrasting sensitivities to absorption (low frequency) as well as reduced scattering (high frequency) coefficients.^32,42,44 In this work, the derived values of optical properties and chromophore responses were considered to be average values resulting from sampling bulk volume. The remitted reflectance was captured simultaneously by the respective channels and saved for processing and analysis off-line using in-house scripts written in MATLAB^®. Prior to camera image and spectrometer raw data analysis, the collected diffuse images at both wavelengths and raw vectors were normalized and filtered to reduce different sources of noise.

2.3.1.

Camera processing

For the respective dataset of each repetition, each individual image is an average of 50 consecutive frames captured by the camera to maximize signal intensity, thereby optimizing signal-to-noise ratio. Thus, four averaged serial images were acquired simultaneously by each camera, one at low frequency (DC) and three phase shifted at high frequency (AC). Each experiment was performed in replicates of 10, resulting in 40 images (10 DCs and 30 ACs) per camera. For each DC image acquired at a specific wavelength, a region-of-interest (ROI) was selected manually by the investigators, and the mean value of this ROI, $R_{\mathrm{d}}(\lambda )$, was calculated using the mean2 function in MATLAB^®. DC illumination was also projected upon a diffuse reflectance standard (white Spectralon, WS-1, Ocean Optics) in the FOV, and ROI and mean value $W(\lambda )$, as previously, were obtained. The normalized reflectance was calculated by subtraction from dark response $D(\lambda )$, and a natural logarithmic function ($-{\log }_{10}$) was obtained to derive the absorption coefficient by

Eq. (1)

$${\mu }_{\mathrm{a}}(\lambda )\propto -{\log }_{10}\left[\frac{R_{\mathrm{d}}(\lambda )-D(\lambda )}{W(\lambda )-D(\lambda )}\right].$$

Dark response, $D(\lambda )$, was characterized by acquiring an image while the light source was turned off, and it is related to the thermal noise of the CCD array. The white surface provides $\sim 100\%$ reflectance and was used to compensate for the spectral shape of the light emitted by the projector and the wavelength-dependent sensitivity of the detector and of the detecting system components. Following 10 repetitions ($N=10$), 10 absorption values per wavelength (${\lambda }_i$) were created, ${\mu }_{{\mathrm{a}}_R}({\lambda }_i)$, and then averaged to obtain a single mean value of absorption

Eq. (2)

$$⟨{\mu }_{\mathrm{a}}({\lambda }_i)⟩=\frac{1}{N}\sum _{R=1}^{10}{\mu }_{{\mathrm{a}}_R}({\lambda }_i)i=\mathrm{1,2}.$$

Once the absorption coefficient at different wavelengths is known and combined with the wavelength-dependent molar extinction coefficients ${ϵ}_i(\lambda )$, metabolic properties ($C_i$) such as perfusion, oxygenation, and chemical content (concentrations of hemoglobin, lipids, water, etc.) can be derived by applying the least-squares solution to the Beer–Lambert law, ${\mu }_{\mathrm{a}}({\lambda }_j)=\sum _{i=1}^{Nc}{ϵ}_i({\lambda }_j)\times C_i$, as commonly employed in tissue optics.⁴⁵ $N_{\mathrm{c}}$ represents the number of chromophores in tissue. This equation shows the absorption coefficient (${\mu }_{\mathrm{a}}$) at a given wavelength (${\lambda }_i$) to be a linear combination of the extinction coefficients ($ϵ$) and their corresponding chromophore concentrations ($C$).

High-spatial frequency reflectance at wavelength ${\lambda }_i$ was recorded in three sequential images at differential 120 deg phase shifts, from which the reduced scattering coefficient image was approximated as follows:

Eq. (3)

$${\mu }_{\mathrm{s}}^{\prime }({\lambda }_i)\propto \sqrt{{[I_1({\lambda }_i)-I_2({\lambda }_i)]}^2+{[I_1({\lambda }_i)-I_3({\lambda }_i)]}^2+{[I_2({\lambda }_i)-I_3({\lambda }_i)]}^2},$$

where $I_1({\lambda }_i)$, $I_2({\lambda }_i)$, and $I_3({\lambda }_i)$ represent the captured reflectance images at spatial phases of 0 deg, 120 deg, and 240 deg, respectively. Similar to absorption, using this relation we mapped light scattering at each wavelength on a pixel level. From this scattering image, an ROI was selected and the mean value ${\mu }_{{\mathrm{s}}_R}^{\prime }({\lambda }_i)$ was calculated (mean2 in MATLAB^®). Following 10 repetitions, a vector of 10 scattering coefficient values was created and then averaged to derive a single reduced scattering coefficient for each wavelength (${\lambda }_i$)

Eq. (4)

$$⟨{\mu }_{\mathrm{s}}^{\prime }({\lambda }_i)⟩=\frac{1}{N}\sum _{R=1}^{10}{\mu }_{{\mathrm{s}}_R}^{\prime }({\lambda }_i),i=\mathrm{1,2}.$$

Please note that by averaging scattering images at each wavelength, the scattering spectra of the sample can be derived via the wavelength-dependent power law function in the form of ${\mu }_{\mathrm{s}}^{\prime }(\lambda )=A_{\mathrm{s}}{\lambda }^{-sp}$, where $A_{\mathrm{s}}$ is the scattering amplitude and sp is the scattering power (slope).^46,47 This functional dependence is supported by numerous studies and is commonly employed for tissue optics.⁴⁸ By analyzing the spectra of scattered light, information regarding macroscopic cellular morphology can be retrieved through the scattering magnitude ($A_{\mathrm{s}}$) and power (sp). In particular, sp is related to mean size of the tissue scattering agents (cell membrane, nuclei, lysosomes, mitochondria, and other organelles in the cytoplasm) and defines spectral behavior of the reduced scattering coefficient, while $A_{\mathrm{s}}$ related to density and distribution of scattering agents, such that the value and sign of sp ($0.2<sp<4$) is a frequently used tool for differential diagnosis of tissue pathologies.^48–50

2.3.2.

Spectrometer processing

Signal processing by the spectrometer is similar to that outlined above, with a few modifications. Once the DC illumination is projected onto the sample, the remitted reflectance from the entire observed area is averaged by LS2, transmitted to the spectrometer through single fiber optic probe, and then recorded, creating a vector comprised of the values from hundreds of wavelengths. In contrast to that described above, Eq. (1) is carried out automatically by the spectrometer software, after which the log operation is performed off-line. Following 10 repetitions, 10 absorption spectra per repetition are recorded ${\overrightarrow{\mu }}_{{\mathrm{a}}_R}({\lambda }_i)$ and then averaged together to obtain a single absorption spectra

Eq. (5)

$$⟨{\overrightarrow{\mu }}_{\mathrm{a}}(\lambda )⟩=\frac{1}{N}\sum _{R=1}^{10}{\overrightarrow{\mu }}_{{\mathrm{a}}_R}(\lambda ),$$

wherein $650\mathrm{nm}\le \lambda \le 1000\mathrm{nm}$, with a spectral resolution of $\sim 2\mathrm{nm}$. To eliminate the high-frequency noise produced by the spectrometer, absorption spectra were then filtered using a second-order Savitzky–Golay filter in MATLAB^®. With the AC illumination, each spectra ($I_1$, $I_2$, and $I_3$) is saved as a vector: ${\overrightarrow{I}}_1$, ${\overrightarrow{I}}_2$, ${\overrightarrow{I}}_3$, and Eq. (3) is performed, creating a scattering coefficient vector over hundreds of wavelengths per repetition ($R$)

Eq. (6)

$${\overrightarrow{\mu }}_{{\mathrm{s}}_R}^{\prime }(\lambda )\propto \sqrt{{[{\overrightarrow{I}}_1(\lambda )-{\overrightarrow{I}}_2(\lambda )]}^2+{[{\overrightarrow{I}}_1(\lambda )-{\overrightarrow{I}}_3(\lambda )]}^2+{[{\overrightarrow{I}}_2(\lambda )-{\overrightarrow{I}}_3(\lambda )]}^2}.$$

At the end of repetitions, 10 vectors of scattering coefficient values at multiple wavelengths are created and then averaged to derive a single vector of the reduced scattering coefficient

Eq. (7)

$$⟨{\overrightarrow{\mu }}_{\mathrm{s}}^{\prime }(\lambda )⟩=\frac{1}{N}\sum _{R=1}^{10}{\overrightarrow{\mu }}_{{\mathrm{s}}_R}^{\prime }(\lambda ).$$

In contrast with the image processing algorithm, the spectral resolution of the scattering spectra dramatically increases, enabling the derivation of tissue structural variations ($A_{\mathrm{s}}$, sp) via the power law equation ($A_{\mathrm{s}}\times {\lambda }^{-sp}$).

It should be stressed that to reduce data acquisition time, thereby allowing rapid extraction of optical properties and speeding up the imaging procedure, the current setup can project only the three AC frames $(I_1,I_2,I_3)$ and obtain the DC ($f_x=0{\mathrm{mm}}^{-1}$) image by the following calculation:

Eq. (8)

$$I_{\mathrm{DC}}=\frac{I_1+I_2+I_3}{3}=\frac{I_o\exp (-0\mathrm{deg})+I_o\exp (-120\mathrm{deg})+I_o\exp (-240\mathrm{deg})}{3}=I_o,$$

given that the DC image is proportional to the absorption coefficient [through the log function, Eq. (1)]. Please note that in the original setup (Ref. 31) a total of six images per wavelength are needed (three phases each for AC and DC) to extract sample optical properties. Recent works aimed toward development of real-time optical imaging based on structured illumination have reported single^51,52 or coupled projection⁵³ images per wavelength. Comparison of the DC image obtained by Eq. (8) with uniform DC illumination on a human hand at 800 nm (Fig. 2) demonstrates good agreement between the two approaches. Also shown in the same figure is the absorption spectrum obtained by Eq. (8) relative to the planar (uniform) illumination. Spectra appear identical, with maximum difference amounting to less than 0.5% (at 990 nm) as presented by the difference plot shown in the same figure. Hence, it may be feasible to extract the absorption spectra or coefficient with Eq. (8), in turn reducing the acquisition and processing time. Nevertheless, since in the current study we were not concerned with real-time imaging, we used the four-projection method.

Fig. 2

(a) AC frequency ($f_x=0.27{\mathrm{mm}}^{-1}$) with phase shift of 120 deg between patterns. (b) DC frequency obtained by Eq. (8). (c) Direct DC (planar or uniform) illumination. (d) Difference image highlights zero discrepancies between the two. (e) Absorption spectrum comparison obtained by Eq. (8) against the DC illumination. The bottom graph marked with the dash-dot line represents the difference between absorption spectrum plots.

3.

Results and Discussion

As mentioned above, calibration and validation experiments were conducted with two different solid-tissue phantoms with known optical properties over the wavelength range of 650 to 1000 nm. In total, a set of two phantoms for calibration and two phantoms for validation purposes was used. Following the calibration process, a scale factor was determined for the respective true values of scattering and absorption coefficients. These factors were used later for estimation of the true optical properties of the validation phantoms and of the human tissue. Representative traces reconstructing the absorption and reduced scattering coefficients at 650 nm (left column) and 800 nm (right column) for validation phantom #1 through camera processing over 10 repetitions during nonconsecutive days ($n=4$) appear in Fig. 3(a). The data points in the graphs represent the mean and the error bars refer to the standard error calculated for the measurements. The error bars for the scattering are not seen because of minute deviation from the mean. A similar trend and average values were observed for validation phantom #2 (data not shown). Stability of the measurements between days of experimentation indicates the minimal influence of artifacts on measurements. The results for the same phantom derived by spectrometer processing at the two specific wavelengths are presented in Fig. 3(b). Also here, the error bars for the scattering are not seen, and overall stability in measurements is observed. The extracted average values from both figures are given in Table 1. A difference in both estimated coefficients between the camera and the spectrometer channels is recognizable, particularly for the scattering coefficient, with a $\sim 33\%$ difference between the two methods at 650 nm. By contrast, less than 5% difference in absorption value between the two channels was found. Figure 4 presents bar graphs summarizing the average and standard error of both extracted optical properties at the two wavelengths as obtained by the camera and the spectrometer channels for validation phantoms, in comparison with their anticipated values. Both figures clearly demonstrate discrepancies between the calculated optical properties and their actual values.

Fig. 3

(a) Representative profiles of the absorption (upper panels) and reduced scattering coefficients (lower panels) at two wavelengths during validation on tissue-mimicking phantom (not included in the calibration process) using the camera channel. $N=4$ represents the mean over four nonconsecutive days. Data points represent the mean, and a bar refers to standard error. Error bars for the scattering are not seen because of minute deviation from the mean. (b) The same as in (a) but reflecting the spectrometer channel.

Table 1

Comparison of recovered average optical properties ± standard error (SE) for absorption and reduced scattering coefficients of phantom #1 between CCD cameras channel and spectrometer channel as extracted from Fig. 3. ± is the root of the sum of the error squares; ∑i=110(SE)i2.

Fig. 4

Quantification of (a) the absorption and (b) scattering coefficients at the two wavelengths obtained by the camera and spectrometer channels following validation analysis of validation phantom in comparison with the actual value. Each bar represents the $\text{mean}\pm \mathrm{SE}$ over ten repetitions during four nonconsecutive days.

Referring to Fig. 4(a), the difference at both wavelengths between the extracted absorption coefficient and their anticipated values for both channels reached 20%, with interchannel deviation of $\sim 5\%$. For the reduced scattering coefficient, Fig. 4(b), an average deviation of $\sim 10\%$ from the actual values was observed by the camera array, while variance of $\sim 40\%$ and $\sim 30\%$ was detected among the spectrometer measurements at wavelengths 650 and 800 nm. Differences between detected and actual values may have several origins: (1) differential geometry employed by the two channels for reflected light collection, (2) aberrations in light collection efficiency, (3) the different methods by which the FOV is selected: with the camera an ROI is selected manually while with the spectrometer the entire FOV is averaged into one single value, unavoidably introducing additional noise, and (4) approximation of the optical properties calculated using Eqs. (1) and (3). Other sources of error and their possible solutions were reported recently.⁵⁶ We believe that such discrepancies can be minimized and that better precision can be obtained in the future by utilizing a larger set of phantoms possessing different optical properties over the spectral band in which the measurements were taken (650 to 1000 nm) to optimize calibration factors. Procedures for correction of sample height- and angle- (surface profile) based structured illumination, suggested previously to recover optical properties more accurately,^52,57,58 will be applied in our ongoing research.

Next, the system was used to measure the optical properties and hemoglobin oxygen saturation of skin tissue of two healthy adult volunteers. Measurements were obtained from the right back hand of subject D (male) and the left back hand of subject M (female); in both subjects, the observed area was the back wrist adjoining the base of the hand. Figure 5(a) shows a photograph of the region captured from subject D, and under DC [Fig. 5(b)] or AC illumination [Fig. 5(c)], alongside plots of the corresponding measured absorption and scattering spectra by the spectrometer [Fig. 5(d)]. A clear decrease in absorption spectrum is evident in the upper panel of Fig. 5(d), which we attribute to differences in concentration of both hemoglobin and melanin. At the same time, two peaks at $\sim 910$ and 980 nm were observed, corresponding to lipid and water absorption, respectively.^59,60 By fitting the skin’s reduced scattering spectra ${\mu }_{\mathrm{s}}^{\prime }(\lambda )$ to a scattering power law ($A_{\mathrm{s}}\times {\lambda }^{-sp}$), we show monotonous spectra decrease, correlating with increased wavelength, such that $A_{\mathrm{s}}=8.4$ (scatter density) and $sp=2.4$ (scatter average size). The sp result is within the range documented by other groups (Table 1 in Ref. 61). As commonly reported, between 700 and 900 nm, absorption by skin chromophores is minimal, while scattering effects are dominant,^62–64 as reported presently. The bar graphs shown in Fig. 6 summarize the average and standard error of both optical properties from the two volunteers (subjects D and M) at the two wavelengths recorded by the camera (upper panel) and spectrometer (lower panel), relative to reference values and values obtained through the diffusion model theory [Eq. (9)]. As expected, the optical properties differ between volunteers as a result of their different BMIs and skin color. As in the phantom experiments, discrepancies were detected between the calculated and reference values: all obtained parameters were lower than reference values, with the exception of scattering at 800 nm in the CCD channel. A possible explanation for these differences may lie in the reference values being averaged from five published studies using different experimental conditions, model samples (skin color, age, etc.), and measuring setup.^63–68 Additionally, partial volume effects may contribute to this discrepancy. Another point that should be taken into account in future experiments with human subjects is the correction of absorption against water concentration in addition to correction sample profile (height and angle mentioned above), which can lead to strong variations in optical properties.^69,70

Fig. 5

(a) Photograph of the right back hand of subject D. Two spatial frequencies that were sequentially projected: (b) DC and (c) AC illumination, respectively. (d) Plots of the measured absorption (upper panel) and reduced scattering spectra (lower panel) from 650 to 1000 nm through the spectrometer processing channel. As expected, the scattering coefficient follows a power law dependence seen in tissue.

Fig. 6

Statistical analysis of the optical properties from two volunteers (subject D and subject M) at the two wavelengths obtained by (a) the camera and (b) spectrometer channels. Comparison of literature-reported values and the model theory [Eq. (9)] are illustrated. Subject D: gender: male, age: 21 years old, BMI: 16.90 (underweight), hand color: black, hand side: right back wrist area. Subject M: gender: female, age: 25 years old, BMI: 27.34 (overweight), hand color: white, hand side: left back wrist area.

In addition, a comparison with the results obtained using recent light propagation models was performed and plotted in Fig. 6. With this model, the total diffuse reflectance measured from a semi-infinite medium using an exponentially decaying source is given as⁴¹

Following the extraction of tissue optical properties, we continued with an estimation of hemoglobin oxygen saturation level (${\mathrm{StO}}_2$). Quantitative assessment of human hemoglobin oxygenation levels is an important tool for monitoring tissue health and for managing a variety of clinical situations. There are several approaches currently in use for evaluating oxygen saturation by tissue spectroscopy. Among them is that defined by the following:⁷¹

Fig. 7

Comparison of tissue oxygenation saturation level between volunteers and between channels. The percent change for each possibility is given.

Hemoglobin oxygen saturation levels can also be derived from the intensity of reflected light; since the wavelength of 800 nm is close to the isosbestic point where oxy- and deoxyhemoglobin absorbs light equally, total hemoglobin concentration (THC) can be readily obtained. In contrast, at 650 nm, the overwhelming majority of absorption is attributed to deoxyhemoglobin (Hbr). Hence, by analyzing the spatiotemporal changes in diffuse reflected light at 650 and 800 nm, the variations in hemoglobin oxygenation may be derived. Since $\mathrm{THC}={\mathrm{HbO}}_2+\mathrm{Hbr}$, saturation level can be calculated by

A representative spatial ${\mathrm{StO}}_2$ image and its histogram distribution derived using this equation for subject M is presented in Fig. 8. Scale values of oxygen saturation are represented by the scale down the image. Averaging the values of the image map revealed ${\mathrm{StO}}_2=0.70\pm 0.062$, amounting to a 9% deviation from that obtained with Eq. (10) (Fig. 7, upper-right).

Fig. 8

(a) Photograph of the left hand of subject M with the dashed rectangle highlighting the selected ROI selected during processing. An example of the absorption and scattering spatial maps at 650 nm within the selected ROI is also presented. (b) Tissue oxygen saturation map of the ROI with a corresponding histogram (c) was generated by Eq. (11). This mapping reveals spatial oxygen saturation heterogeneities within the measured tissue. The horizontal color bar represents the saturation value of each pixel in the map, such that higher saturation values correspond to brighter pixels. Conversion of the saturation map into histogram distributions demonstrated localized changes in the saturation levels at single-pixel resolution and illustrates the distribution of oxygen saturation within the FOV. The vertical axis in the histogram reflects the number of counts in each bin, and the solid curve is a Gaussian fit with mean and standard deviation of 0.70 and 0.062, respectively. Both map and histogram illustrate the fine resolution at which the system can image tissue oxygen saturation parameters.

4.

Summary

In this study, a noncontact optical setup integrating spectrometer and camera arrays was established to quantify optical properties of both tissue-phantoms and biological tissue over a large FOV by spatial light modulation (structured illumination). In this setup, sinusoidal light patterns are serially projected onto the sample at both low and high spatial frequencies to isolate the target’s absorption (linked to tissue metabolism) and scattering (related to tissue structure) properties. The diffuse reflected light is simultaneously acquired by single spectrometer and passes through two parallel narrow bandpass filters, each leading to an acquisition camera. In addition to the extraction of the tissue’s optical properties, we calculated hemoglobin oxygen saturation levels from the hands of two healthy volunteers (Fig. 7). A major advantage of this parallel optical configuration lies in the ability of each component to complement the other, enabling high spectral and spatial resolution. Since this combined platform provides complementary information both spectrally and spatially, it was able to detect variance between the optical properties recorded by the camera and spectrometer channels, as well as between the recovered absorption and scattering coefficients and those expected (Fig. 6). An additional validation methodology based on the theoretical model-based diffusion equation further revealed discrepancies between the calculated and theoretical values. The discussion raised several factors that may account for such differences, together with proposed adjustments to mitigate these potential artifact sources, as part of our ongoing research. We are currently optimizing strategies to further improve the system’s performance. The present work is not the first to demonstrate wide-band tissue optical spectroscopy utilizing sinusoidal patterns of spatially modulated light.⁷⁵ However, to the best of our knowledge, there is no example of such combined spatially modulated optical setup for tissue optical characterization.