This PDF file contains the front matter associated with SPIE Proceedings Volume 13011, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

This study introduces an innovative chip-upending technique to enhance the quality of silicon nitride waveguides by minimizing sidewall roughness, which is critical for reducing optical propagation losses. By reflowing the resist after development, we effectively control the resist expansion effect typically observed in standard procedures. Our analysis indicates that this method can decrease line edge roughness (LER). We employ atomic force microscopy (AFM) to measure the LER of the resist and the waveguide sidewalls via a special tilting technique, ensuring precise characterization of the surface topography. The fabrication is carried out by the choice of a suitable DOE for ensuring the statistical robustness in the evaluation process. Additionally, we conduct propagation and bend loss measurements at 850 nm across waveguides of various widths, with thicknesses ranging from 200 to 400 nm. These measurements in correlation with the roughness analysis confirms that sidewall roughness is indeed smoothed, resulting in notable improvements in light propagation efficiency. The versatility of the low-temperature reflow process is further demonstrated by its compatibility with several waveguide geometry alterations . The inclusion of tree-based modeling in optimizing relevant parameters in the fabrication, foremost the reflow parameters, thereby contributing to a more efficient and controlled reflow process. The findings suggest that our approach can be universally adopted for the fabrication of low-loss waveguides in photonic integrated circuits, with necessary wafer-to-wafer stability.

Accurate measurement of small RIU, surface refractive index (RI) changes is crucial in biosensing. This study introduces a novel approach to correlate intricate spectral features with RI variations in Surface Plasmon Resonance (SPR) sensing using optical fiber gratings. Using a regression model with gold-coated tilted fiber Bragg grating (Au-TFBG) sensors, we achieve enhanced signal stability and precision across diverse experimental setups. This eliminates the need for sensor calibration, streamlining biosensing protocols. Our findings represent a significant advancement in real-time RI monitoring, offering promising applications in biosensing.

Energetic charged particles from the secondary cosmic radiation may interact with Raman measuring devices, leading to local charge anomalies in the signal from the CCD detector represented as single to few samples wide spikes in the resulting Raman spectrum. If not handled properly, these spikes may lead to errors in subsequent data processing steps. There are no practical means to shield the energetic muons, thus spike removal is a common and fundamental step in Raman spectral data pre-treatment. Commercial software from Raman instrument manufacturers usually provide spike removal tools, but the algorithms are not available to users and automation or control of the corrections is impossible. Open-source tools for spike removal are also limited. Moreover, the performance of these tools is rarely validated or benchmarked. The present work is an introduction to a bigger study which aims to perform a systematic comparison of spike removal algorithms. The methodology consists of artificially introduced spikes on a spike-free spectrum. Minimum detectable spike amplitude for four different algorithms is studied by performing a scan, varying the location and amplitude of the artificial spikes. Studied algorithms are made available in RamanChada2 library.

The automation of spectral classification tasks has made machine learning models essential analytical tools. However, the complexity of hyperparameter tuning limits the practical use, particularly for novices. This study applies these classifiers to identify bacteria using surface-enhanced Raman spectroscopy (SERS), offering a rapid and non-invasive alternative to the gold standard, matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF-MS). An evolutionary algorithm was employed to optimize the hyperparameters of 10 machine learning models. We found the topperforming model for the classification of the SERS spectra of E. coli and S. pneumoniae water suspensions. This approach yielded a test accuracy of 95.8%, 100%, 100% when using the Bernoulli Naïve Bayes, Support Vector Machine, and Multilayer Perceptron models, respectively. This demonstrates the potential of self-optimizing machine learning models as accessible analytical tools for diverse classification tasks in biophotonics. This automated approach extends to identify various samples and data structures, not just pathogens’ spectra.

The rise of materials informatics and artificial intelligence (AI)-based computational discovery of materials provides new sources of research directions for materials science. New research demonstrates the enormous potential for experimentalists in photonics for photovoltaic materials to increase the rate of screening and optimization of materials properties and related devices. AI for materials is not only interested in the accuracy of predictive models but also in the effect of data size. Recent investigations have shown significant progress in AI for small data by combining a design of experiments (DoE) approach with machine learning (ML) analysis, which enables experimentalists to use scarce resources more effectively for materials optimization with a higher probability of arriving at a true optimum.

In this work, we propose an alternative approach to DoE associated with ML by using the concept of active learning (AL). AL is well appropriate in industry and physical sciences, where there is a strong need to minimize the number of costly experiments necessary to train predictive models. We focus on optimizing processes of organic photovoltaic (OPV) cells. The manufacturing of OPV devices requires on the case of having a very small labeling budget, about a few dozen data points, and developing a simple and fast method for practical AL with a model selection. Then, we discuss the challenges in anticipating the data-driven process design, such as the complexity of the experimental approach of OPV cells, the diversity of experiment parameters, and the necessary programming ability.

Raman spectroscopy, a non-invasive analytical method, offers insights into molecular structures and interactions in various liquid and solid samples with applications ranging from material science, and chemical analysis to medical diagnostics. Preprocessing of Raman spectra is vital to remove interferences like background signals and calibration errors, ensuring precise data extraction. Artificial intelligence, particularly machine learning (ML), aids in extracting valuable information from complex datasets. However, effective data preprocessing proves to be crucial as it can influence model robustness. This study addresses the integration of preprocessing and ML algorithms, often treated as distinct identities despite their intrinsic interconnection, in Raman spectra of blood samples from patients suffering from ovarian cancer. Optimal preprocessing configuration may not always be evident due to the complexity of spectral data. There are numerous options available for background corrections, normalization, outlier removal, noise filtering, and dimension reduction algorithms for Raman spectra. Moreover, hyperparameter tuning is required to detect the best choices for the preprocessing steps. In this work, we present a pipeline to co-optimize preprocessing techniques and ML classification methods to promote objective selection and minimize processing time. In our approach, preprocessing methods are not chosen arbitrarily but rather systematically evaluated to enhance the robustness of the models. These criteria focus on ensuring that the model performs well not only on the training data but also on unseen data, thus reducing the risk of overfitting and improving the generalization capability of the model. This systematic approach would reduce the time for new studies by detecting the most suitable preprocessing steps and hyperparameters needed and building a robust model for the task.

In the realm of ex-vivo diagnosis, the integration of optical and chemical imaging data has emerged as a transformative approach, offering a comprehensive understanding of biological specimens at a molecular level. Chemical imaging of human tissue specimens provides an all-digital label-free approach to imaging in objective histopathology, though it requires reference to gold standard pathological (e.g. haematoxylin and eosin (H+E) stained) images for pathological interpretation.

Optical imaging techniques, such as microscopy and spectroscopy, provide detailed spatial information, capturing morphological features with high resolution. Concurrently, chemical imaging methods, including mass spectrometry and Raman spectroscopy, offer insights into molecular composition. The challenge lies in harnessing the complementary strengths of these disparate modalities to extract a holistic understanding of the sample.

In this work we present the results of several image alignment approaches for fusion and integration of chemical and pathological imaging data, demonstrating that the process of corner detection is crucial towards precise image alignment.

The FRD-PVOH algorithm/medical device enables continuous noninvasive spectroscopic analysis and monitoring of physiology induced changes in the peripheral vasculature in vivo in humans, specifically red blood cell and plasma volume fractions. Previously, FRD-PVOH demonstrated a correlation between mean arterial pressure (MAP) and vascular plasma volume with tilt-table experiments, indicating physiological relevance. Continuous monitoring of plasma volume and hematocrit (Hct) with a total bandwidth of 0-0.3 Hz in volar side fingertip capillary beds, reveals MAP fluctuations superimposed on slower background. These MAP fluctuations isolated from slower vascular volume shifts and the cardiac pulse create a time series associated with vasodilation/vasoconstriction i.e., thermoregulation, an autonomic path to homeostasis. We consistently observe random walk dynamics i.e., the amplitudes of the fluctuations in the time domain are distributed as a Gaussian while Fourier analysis simultaneously confirms the presence of circulatory physiology in the results i.e., breathing effects and the Baroflex response. Lorenz plots indicate linear dynamics control short and long timescale fluctuations across both test subjects. ANOVA analysis of data from a small, on-going, study using a mild external thermoregulation protocol (2 supine subjects, 16 total hours of monitoring for each subject, in 8 sessions of 2 hours each) demonstrates that the number of fluctuations per unit time varies across the two subjects with 95% significance with or without the thermoregulation challenge. Also, the mean number of fluctuations per monitoring session was significantly greater for sessions that included an external thermoregulation challenge compared to those that did not, for both subjects at 95% confidence.

Monitoring the quality of extra virgin olive oil (EVOO) during its life cycle is of particular importance due to its influence on health-related characteristics and its significance for the oil industry. For this reason it is critical to find an easy-to-perform, non-destructive and affordable method to monitor the quality of EVOO and detect its degradation due to ageing. The following study explores a machine learning approach based on fluorescence measurements for predicting oil changes arising from the ageing process. The proposed method specifically predicts the quality parameters that are required for an olive oil to qualify as extra virgin. In particular, the two properties considered in this analysis are the UV absorbance at 232 and 268 nm (K₂₃₂ and K₂₆₈), both critical markers of the quality of extra virgin oil. To achieve this goal, a large dataset of fluorescence measurements was analysed, comprising 720 excitation-emission matrices of twenty-four different oils initially labeled as extra virgin. The samples were aged under accelerated conditions at 60 °C in the dark for nine weeks and their properties were measured at ten different time steps during the process.

Revision total hip arthroplasty suffers from low visibility with intra-body navigation hinging primarily on auditory and tactile cues. Consequently, the risk of surgical injury increases. One proposition to increase surgical precision is integrating an algorithm which classifies encountered tissues based on their reflectance spectra into the surgical tools. Previous works have developed machine learning applications for the automatic, binary, classification of tissue based on diffuse reflectance spectroscopy (DRS) signals and exploratory investigations have successfully integrated DRS probes into surgical devices including surgical drills. However, one problem with these studies is a lack of transparency in the algorithms, which is important to increase practitioners’ trust and prevent bias. This study developed four machine learning algorithms which simultaneously classified broadband DRS signals (355 – 1850 nm) of six ovine tissue classes. The algorithms were Linear Discriminant Analysis (LDA), Random Forrest, Convolutional Neural Network (CNN), and a Transformer model. Class-wise wavelength importance was visualized using model-based methods to understand classification mechanisms and increase model-explainability. It is concluded that CNNs hold the potential for successful initial device design and medical integration.

Recent strides in data-driven and deep learning methods have empowered image and wavefront reconstruction in such environments. This breakthrough finds promising roles in biomedical applications like image transmission and holography. Yet, the reconstructed image quality relies on deep learning model effectiveness in understanding transmission mechanisms. In our presentation, we propose two enhancements. First, employs a novel deep learning architecture inspired by light physics, showcasing enhanced image reconstruction quality and broad problem generalization. The second one is an optical method which boosts data variance through holographic encoding, enabling multi-channel image transmission and improved data fusion via deep learning.

The rapid advancement of imaging technologies in pathology has ushered in an era of data-intensive diagnostic workflows, generating large volumes of data that demand sophisticated segmentation and compression techniques. Chemical imaging approaches offer an all-digital objective approach to pathological analysis, though image segmentation is required for efficient computation.

Convolutional autoencoders are highly connected deep learning networks which can learn salient features within imaging data for the purposes of compression, data recovery, development of classifiers and/or segmentation.

In this study an objective analysis of a U-Net convolutional autoencoders for unsupervised image segmentation is conducted with respect to haematoxylin-eosin based ground-truth diagnostic pathology. We find that a light-weight network architecture may provide a suitable segmentation approach for chemical imaging.

Clinical pathological diagnosis and prognosis for cancer is often confounded by spatial tissue heterogeneity. This study investigates the utility of entropy as a robust quantitative metric of spatial disorder within Fourier Transform Infrared (FTIR) chemical images of breast cancer tissue. The use of entropy is grounded in its capacity to encapsulate the complexities of pixel-wise spectral intensity distributions, thus providing a detailed assessment of the spatial variations in biochemistry within tissue samples.

Here we explore the use of Shannon’s entropy as a single image-based metric of spectral biochemical heterogeneity within FTIR chemical images of breast cancer tissue. This metric was then analyzed statistically with respect to hormone receptor status. Our results suggest that while entropy effectively captures the heterogeneity of tissue samples, its role as a standalone predictor for diagnostic subtyping may be limited without considering additional variables or interaction effects. This work emphasizes the need for a multifaceted approach in leveraging entropy with chemical imaging for diagnostic subtyping in cancer.

The identification of spectral markers differentiating disease states when using spectral data is challenging in the context of modelling with deep neural networks, particularly in scenarios where classification models are developed with multiple classes.

While a number of approaches do exist which can provide an insight into the features which are learnt by deep learning models, in biophotonics and chemical imaging these have received relatively little attention. In the present work we pilot the use of Fourier Transform chemical imaging with two deep-learning interpretation approaches within the context of a multi-class classification problem. Fully connected neural networks are developed on unfolded chemical imaging data captured on patient-derived xenografts developed from a colorectal cancer model. Separately, Shapley additive explanations and saliency approaches are used to derive feature sets which are discriminatory for class within this experimental model of colorectal cancer.

Preliminary results suggest that Shapley additive explanations provide differentiating spectral sets which may not be derived with saliency, although the feature sets which are identified are dependent upon spectral pretreatment methodology. A dual approach which employs both strategies may be an effective strategy for the identification of feature sets in this context.

Perovskite solar cells (PSCs) are renowned for their efficiency, affordability, and mass manufacturing. However, the performance unpredictability, material sensitivity and stability issues, and optimization limit their practicality. This study includes the challenges related to PSCs and the role of Artificial Intelligence (AI) in their advancement. AI has shown that it can accelerate the PSC's designs by finding creative solutions. The design assistance provided through AI-based methods reduces the experimentation time and need for resources, enabling real-time production monitoring and control. These methods identify performance bottlenecks and forecast the device efficiency in various settings. In this paper, we have simulated three perovskite solar cell devices (MASnI₃, FASnI₃, and MAGeI₃) using SCAPS-1D with ETL as ZnO and HTL as Cu₂O. Random Forest technique has been used for optimization and prediction of the best PSCs efficiency where the conduction band density of state, thickness of the absorber layer, hole mobility, valence band density of state, and electron mobility have served as design variables. The MSE and R² scores for performance prediction are 1.37× 10^-3 and 0.992 for MASnI3, 4.21 × 10^-3 and 0.997 for FASnI₃, and 0.79 × 10^-3 and 0.993 for MAGeI₃ respectively.

Metasurfaces have been emerging increasingly due to their realization of various technologies in meeting the design of multi-functional, compact, highly efficient, tunable, and low-cost designs owing to the fact that they can manipulate electromagnetic (EM) waves in a sub-wavelength thickness. In the optical regime, they have been successful in realizing transmission, reflection and absorption for a wide range of interesting applications. The metasurface absorbers have found place in energy harvesting applications. However, their design and analysis is carried out using EM solvers which in general are heavily time-consuming due to their iterative nature of solving a problem. To mitigate the problem of slackness and computational burdensome, the machine learning (ML) is becoming popular for tackling the data related problems and have been in use for making the design of metasurfaces faster. In this work, three ML algorithms namely, XGBoost, Least Absolute Shrinkage and Selection Operator (LASSO) and Random Forest (RF) have been applied both in forward and inverse topologies for a tungsten based square-ring meta-absorber. The inverse training has been carried out by employing “principal component analysis” (PCA). The operation of a meta-absorber is dependent on its geometry; thus, the training has been carried out by varying all the geometrical features of the unit element under study. The prediction performance of the presented regression models is reckoned to be accurate that the predicted values are in the near vicinity of ground truth values. The minimum MSE for the forward model attained for the case of RF is 5.08 ×10⁻³ and that of R² is 0.9952, whereas for the inverse model, the minimum MSE of 2.05 and R² score of 0.958 with 200 PCA components is achieved. The prediction time is minimum for the LASSO algorithm which is as low as one second. The lower computation time, reliable prediction, and model-free nature of ML techniques have made them useful against data imperfections and are proven to be an effective solution to time-consuming and computationally expensive tools for metasurface design.