Significance:White blood cells (WBC) are hematopoietic cells of the immune system that protect the body by recognizing and eliminating infectious agents. Abnormalities in WBC production, maturation, or function can lead to disease and associated morphologic changes that, when systematically characterized, support diagnostic classification and clinical decision-making. Aim:We aim to investigate polarized hyperspectral imaging (PHSI) and polarized light imaging (PLI) microscopy for the visualization of WBCs. Approach:We developed a dual-modality microscopic imaging system that performs both polarized hyperspectral imaging and polarized light imaging. In the dual imaging setup, we used a snapscan hyperspectral camera and an RGB camera to acquire images separately and further calculate four Stokes parameters (S0, S1, S2, and S3) as well as three Stokes vector-derived parameters, namely, the degree of polarization, degree of linear polarization, and degree of circular polarization. Synthetic RGB images of Stokes vectors and Stokes vector-derived parameters were generated for the visualization of cellular components with PHSI images. The spectral signatures of representative WBCs, e.g., granulocytes and lymphocytes, were extracted for qualitative comparison. Results:The preliminary results demonstrate that Stokes vector parameters can enhance the visualization of granules in granulocytes, the visualization of surface structures of lymphocytes, and the morphologic visualization of the monocyte nucleus. Furthermore, the results also reveal that the measured spectra of Stokes vector parameters could enhance the differentiation of WBCs in the spectral dimension, represented by the qualitative comparison between granulocytes and lymphocytes. Conclusions:Utilizing the spatial and spectral information from the Stokes vector data, our customized polarized hyperspectral microscopic imaging system enhances the visualization of WBCs and may provide a tool for the diagnosis of disorders related to white blood cells.

Digestive system cancers, particularly esophageal cancer, remain difficult to detect in early stages due to the limited sensitivity of conventional white-light endoscopy. Hyperspectral endoscopic imaging enhances tissue contrast by exploiting wavelength-dependent optical properties. However, its performance depends on efficient and uniform spectral illumination under highly constrained and asymmetric endoscope tip geometries. In practice, the imaging sensor is offset from the center due to the existence of instrument channels and air/water nozzles, complicating LED placement and degrading illumination uniformity. In this work, we present an optical simulation-driven framework for optimizing LED placement in an LED-based, asymmetric hyperspectral endoscopic illumination system. Non-sequential ray tracing and merit-function-based optimization were implemented in Zemax OpticStudio to optimize LED spatial configurations. Two local optimization algorithms-Damped Least Squares (DLS) and Orthogonal Descent (OD)-were systematically evaluated across fifteen wavelength groups. Both methods achieved measurable improvements in irradiance distribution. Overall, OD slightly outperformed DLS, yielding average irradiance improvements of 11.30% on the esophagus wall and 10.15% on the LED field of view relative to the pre-optimization configuration, whereas DLS achieved corresponding average increases of 7.03% and 6.55%, respectively. Rather than prescribing a fixed LED placement pattern, the results provide practical design guidance by demonstrating how adaptive, optimization-driven LED positioning can effectively compensate for asymmetric endoscope geometries while prioritizing different illumination targets. These findings demonstrate that simulation-based LED placement optimization can provide effective guidance for the design of hyperspectral endoscopic illumination systems, high-fidelity modeling, and experimental validation.

BACKGROUND AND PURPOSE:IDH mutation & 1p/19q codeletion are critical biomarkers for glioma diagnosis & therapy. 1p/19q codeletion occurs exclusively in IDH-mutated gliomas. In this study, we developed a 2-stage, non-invasive, MRI-based deep learning method that leverages IDH status to enhance 1p/19q predictions. MATERIALS AND METHODS:Multi-contrast brain tumor MRI & genomic information were obtained from five publicly available (TCIA, UCSF, EGD, UPenn & LGG), and three in-house/collaborator institutions (UTSW, NYU, UWM). Subjects were screened for the availability of IDH & 1p/19q status as well as T1, T1CE, T2, FLAIR MR images. For training purposes, missing T1 and FLAIR contrasts for the LGG database were generated using an in-house multi-contrast simulator. Two separate U-Nets were developed for 1p/19q-classification: a multi-contrast network (MC-Net) and a T2w-only network (T2-Net). A separate U-Net was developed for IDH classification (IDH-net). A total of 2044 subjects were used in training and testing IDH-Net, and 1426 subjects were used in training and testing the 1p/19q models. The IDH-Net was trained using subjects from TCIA, UTSW, and UPenn. The 1p/19q networks were trained using subjects from TCIA, UTSW, and LGG. The trained networks were tested on true held-out cases from NYU, UWM, EGD, and UCSF. In the 2-stage approach, subjects were initially classified for IDH status using IDH-Net. Predicted IDH-wildtype cases default to 1p/19q non-codeleted. Then the IDH-mutated cases were further classified for 1p/19q status using the 1p/19q-networks. RESULTS:IDH-Net achieved a classification accuracy of 93.7%. 1p/19q MC-Net & T2-Net achieved classification accuracies of 86.5% & 86.0%, respectively. In the 2-stage approach, 1p/19q MC-Net and T2-Net achieved accuracies of 91.5% & 91.2% respectively, improving the classification accuracy by ∼5%. CONCLUSIONS:This study demonstrates the effectiveness of leveraging IDH status to enhance 1p/19q classification. A ∼5% increase in classification accuracy was achieved when using the 2-stage approach, using IDH-Net to gate 1p/19q predictions. The developed method offers a reliable, non-invasive approach to determine important biomarkers for glioma diagnosis.

Hyperspectral imaging (HSI) has been used in vivo to estimate the oxygenation level (SO2) of whole blood inside blood vessels. However, little literature is devoted to validating the estimation. The goal of this research is to compare the performance of three commonly used methods: the Kubelka-Munk algorithm, optical density algorithm, and inverse Monte Carlo method, for the task of estimating SO2 from diffuse reflectance spectra. Validation was performed using two data types: Monte Carlo simulated data and ex vivo sheep blood that was deoxygenated using sodium dithionite (Na2S2O4). The SO2 of sheep blood was confirmed using a blood gas analyzer. We measured the diffuse reflectance using two types of HSI cameras: a push-broom and a snapshot HSI camera system. Our results showed that visible wavelengths (400 - 600 nm) are preferred for micro-vessels with diameters less than 100 microns, and near-infrared wavelengths (600 - 900 nm) are preferred for larger vessels. We determined experimentally that sodium dithionite consistently deoxygenated whole blood by 35% for every 1 mg/mL of sodium dithionite to hemoglobin. It was found that the inverse Monte Carlo method performed well in estimating SO2, with an average estimation error of 5.53% and 6.97% for ex vivo sheep blood and Monte Carlo spectra, respectively. Our findings provide a basis for determining future methods for in vivo SO2 estimations, including surgical HSI applications that rely on accurate perfusion assessment.

Significance:Early detection of Alzheimer's diseases, diabetic retinopathy, or macular degeneration with advanced retinal imaging technologies can help improve patient care and treatment outcome. Aim:We aim to create a high-resolution hyperspectral imaging (HSI) system for the retina. Retinal vessel diameter and oxygenation rate will be extracted simultaneously from HSI data. Approach:Our hyperspectral retinal imaging system consists of a snapshot hyperspectral camera, a high-resolution RGB camera, a beamsplitter, and an imaging endoscope. Multiple pansharpening algorithms, including deep learning methods, were developed to generate high-resolution hyperspectral images that were further used for the measurement of vessel size and oxygenation rate in mice. Results:The hyperspectral retinal imaging system was tested for its spatial resolution and spectral fidelity in retina phantoms. In vivo imaging experiments were performed in mice. The deep learning-based pansharpening algorithm achieved a root mean square error (RMSE) of 2.15 ± 0.64 , a correlation coefficient (CC) of 0.96 ± 0.05 , a spectral angle score of 0.06 ± 0.03 radians, and an error relative global dimensionless synthesis (ERGAS) score of 2.37 ± 1.71 . Oxygen saturation ( sO 2 ) and lumen diameters of blood vessels were measured in the retina. The average lumen diameter of the venules was 45.7 ± 13.6    μ m , whereas the average lumen diameter of the arterioles was 31.5 ± 8.7    μ m . The average arteriole sO 2 was 98%, whereas the average venule sO 2 was 58%. Conclusions:A high-resolution hyperspectral imaging system was developed and validated for retina imaging and measurement of blood vessels and oxygen saturation.

Chemical exchange saturation transfer ( CEST) contrast MRI provides unique molecular contrast by probing proton exchange between solutes and bulk water, but its clinical translation is limited by long acquisition times required to acquire finely sampled Z-spectra bringing a tradeoff between acquisition time and frequency offset responses. In this preliminary study, we investigated the performance of transformer-based deep learning models to reconstruct fully sampled CEST spectra. We first validated the approach using a lightweight model on simulated Z-spectra, followed by a scaled high-capacity transformer for human MRI data to handle the increased data scaling and spectral variability. We evaluate the model reconstruction quality on simulated Z-spectra and the human MRI data using root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and cosine similarity for regression analysis. We use structural similarity index measure (SSIM) and peak signal to noise ratio (PSNR) to evaluate image-level reconstruction quality. The transformer approach accurately recovered missing spectral information with an MAPE of 3.41% and RMSE of 0.0169. The reconstruction maintained high structural fidelity with a mean SSIM of 0.9658 and spectral fidelity with Cosine Similarity of 0.9951. These results validate that the scaled transformer architecture ensures robust generalization and accurately captures complex spectral dependencies in clinical data. These findings highlight the potential of deep learning-based reconstruction to accelerate CEST MRI acquisition supporting more efficient clinical protocols.

Polarized hyperspectral imaging (PHSI) integrates spectral and polarization contrast to enable comprehensive tissue characterization. In this study, we validate a refined handheld PHSI probe that is more compact, lightweight, and capable of rapid image acquisition. The system was evaluated on freshly excised tissues from five male mice, focusing on four organ types-brain, tongue, thyroid, and testicle-chosen for their similar macroscopic appearance but distinct biochemical and microstructural properties. Spectral classification was performed using a one-dimensional convolutional neural network (1D-CNN) trained on total reflectance (Stokes S0) spectra. In parallel, polarization-resolved parametersdegree of linear polarization (DoLP) and degree of circular polarization (DoCP)-were derived to visualize structural and scattering differences among tissues. Despite the visual similarity among tissues, the model achieved high classification accuracy with an average overall accuracy of 95.3% across leave-one-out cross-validation folds. DoLP effectively differentiated glandular from adipose tissue, and highlighted muscle fiber orientation differences within tongue regions. DoCP distinguished blood-rich regions from surrounding glandular or fatty tissues, revealing higher DoCP in vascular areas and lower values in fat. These results confirm that spectral and polarization features offer complementary tissue contrast. This work demonstrates the potential of handheld PHSI for real-time tissue assessment and lays the foundation for future integration of spectral and polarization data in machine learning for clinical decision support.

Hyperspectral imaging (HSI) is an emerging modality that captures rich spatial and spectral information, offering promise for tissue characterization and disease diagnosis. In the context of the prostate, where histopathological assessment remains the gold standard, current imaging techniques often lack sufficient ability to support fine-grained tissue discrimination. Moreover, large annotated HSI datasets are scarce due to the complexity of expert labeling, limiting the development of data-driven diagnostic models. We present a framework for HSI-based analysis of gross prostate datasets, integrating remote annotation for prostate tissue segmentation. HSI data were acquired ex vivo from prostate specimens after radical prostatectomy (RP). Three snapshot HSI cameras were utilized for data acquisition, including a visible (VIS), red/near-infrared (RNIR), and near-infrared (NIR) HSI camera. Digitized histological images were matched to corresponding hyperspectral images. A web-based tool was enabled to facilitate remote annotation on whole-slide images. A spatial-spectral transformer architecture was developed to segment different tissue of the prostate. Our preliminary results show that RNIR achieved the pixel accuracy of 0.694 and NIR yielding a Dice similarity coefficient of 0.601, suggesting that the results can be further improved and that HSI may have potential value in the NIR range. This pilot study provides a pipeline for hyperspectral imaging of human prostate tissue with histologic correlation and validation.

Medical image editing enables simulation of rare disease presentations and counterfactual examples but often struggles to preserve subject-specific anatomy. We propose a zero-shot, prompt-driven framework for anatomically consistent editing of three-dimensional (3D) multi-sequence brain MRI using pretrained latent diffusion models. Our method introduces a feature injection strategy, where intermediate features from residual and attention blocks of a guidance image are incorporated into the denoising process alongside text or mask conditions. This design preserves local structural fidelity and global contextual coherence during editing. We demonstrate three applications: (1) insertion of brain tumors into healthy brains, (2) manipulation of IDH molecular status with mask- and text-guided prompts, and (3) tumor removal with anatomically coherent reconstruction of healthy tissue. Results show anatomically faithful outputs that maintain patient-specific brain structure while performing semantically precise edits. The framework is broadly applicable for simulating disease subtypes, augmenting training datasets, and generating counterfactuals in medical imaging research.

Polarized hyperspectral imaging (PHSI) combines wavelength-resolved reflectance with polarization sensitivity to probe tissue absorption, scattering, and microstructural organization in a label-free manner. We developed a compact handheld PHSI probe that integrates two liquid crystal variable retarders (LCVRs), a linear polarizer, and a hyperspectral snapshot camera to achieve full Stokes imaging over the range of 460-600 nm. In this work, we establish a comprehensive calibration and validation framework for this probe to improve its quantitative spectral and polarization performance. Individual LCVRs were first calibrated at four wavelengths (480, 530, 580, and 633 nm) using a crossed-polarizer method to obtain voltage-retardance characteristic curves and to determine the driving voltages corresponding to 0, lambda/4, and lambda/2 retardance at 530 nm. We then adapted a full-Stokes polarimeter calibration method to identify systematic errors at 520 nm, using six calibration samples and a Stokes-Mueller model that account for LCVR axis offsets, residual retardance errors, and polarizer misalignment. Spatial and spectral performance were evaluated using a USAF target, a ruler, and eight color tiles with known reflectance spectra, resulting in a field of view of 7 mm x 11 mm, a resolution of 7.13 lines per millimeter (lp/mm), and spectral angles below 5 degrees. Additional validation on Intralipid-dye phantoms and ex vivo mouse tissues demonstrated that the calibrated probe could capture dye- and tissue-specific reflectance and degree of polarization spectra. These results highlight the potential of the calibrated handheld PHSI probe as a portable tool for quantitative tissue characterization in biomedical research and diagnostics.

Hyperspectral imaging (HSI) provides high-resolution spectral information capable of distinguishing subtle physiological differences across tissue types, offering significant promise for surgical guidance and point-of-care diagnostics. However, the high dimensionality and computational demands of HSI data have posed barriers to its real-time deployment on lowpower embedded platforms. In this work, we develop an end-to-end, deployable pipeline for tissue-level HSI classification on a Raspberry Pi 5 single-board computer, enabling fast and memory-efficient inference without the need for specialized accelerators. A dataset comprising of 630 hyperspectral images (i.e., hypercubes) of ex vivo abdominal tissues across six organ types and four animal models were curated using snapshot HSI cameras. Our approach uses a ResNet-18 model pretrained on ImageNet dataset and modified to accept 16-channel inputs, was fine-tuned to perform multi-tissue classification, achieving a top 1 accuracy of 85.26% using full-precision (FP32) weights. The trained models were then exported to TorchScript and subjected to post-training dynamic quantization. For the INT-8 ResNet-18 model, this resulted in a fourfold reduction in parameter memory (from 44.9 MB to 11.2 MB) and a substantial decrease in peak RAM usage on Pi 5, from 25.15 MB to 6.31 MB, with no loss in classification accuracy. Benchmarks conducted on the Raspberry Pi 5 CPU indicate that the quantized INT-8 model achieves a mean inference latency of 149.42 ms per 128 x 128 x16 hyperspectral data cube, corresponding to a throughput of 6.37 hypercubes per second. These results demonstrate that high performance HSI classification can be achieved on an embedded hardware costing less than $200, using widely available convolutional architectures and quantization techniques. The proposed pipeline thus offers a practical, accessible, and compact solution for deploying HSI analytics in clinical settings, particularly promising for real-time surgical guidance and point-of-care applications.

Significance:Hyperspectral imaging (HSI) is an advanced spectral imaging technique that captures spatial and spectral information across numerous wavelength bands. This capability allows tissue characterization, disease detection and diagnosis, surgical guidance, and digital histopathology, making it an increasingly valuable tool with wide biological and medical applications. Aim:We aim to provide readers with (1) an understanding of the principles and technological advancements in HSI, (2) a comprehensive overview of HSI data processing and analysis methods, and (3) an updated survey of biomedical applications, from disease detection, intraoperative imaging, to histopathology. Approach:A systematic literature search was conducted using PubMed and Google Scholar with the keyword "hyperspectral imaging." We previously published a comprehensive review paper on medical HSI in 2014, which was widely cited in the field. Therefore, this updated review focused on new technology advancements and emerging applications. Based on their biological and medical relevance, 612 HSI papers were included and analyzed in this review. Results:Recent advances in HSI span both hardware and computational techniques, including improvements in sensor technology, data processing and analysis, short-wave near-infrared imaging, and deep-learning and AI tools. HSI is actively explored for various applications in oncology, neurology, ophthalmology, dermatology, cardiology, gastroenterology, hepatology, wound care, endocrinology, dentistry, infectious disease, plastic and reconstructive surgery, general surgery, intraoperative guidance, histopathology, microbiology, nanopathology, and pharmacology. Conclusions:HSI has become an emerging imaging modality in biomedical research and clinical settings. Continued advancements in hardware miniaturization, computational efficiency, and clinical validation will further solidify the role of next-generation HSI in biomedicine.

Hyperspectral imaging (HSI) can capture spatial and spectral data of the retina. Retinal vessel segmentation enables quantitative analysis and assessment of retinal health. In this work, we developed an HSI system for mouse retina and a deep learning architecture to segment retinal vessels. Retinal images were manually segmented to establish ground truth. We proposed the use of multi-head attention block for the spectral data, as well as Tversky loss for segmentation refinement. The model's performance was assessed using Dice similarity coefficient (DSC) and the Jaccard (IoU) metrics. A comparative analysis was conducted against the manual segmentation, and five other segmentation methods: UNet, UNet++, UNet3+, segment anything model (SAM), and SAM2. We demonstrated that our spectral network achieved an IoU and DSC scores of 0.80 +/- 0.04 and 0.89 +/- 0.02, respectively. Our network outperforms other networks. The automatic segmentation method provides a tool for retinal analysis and quantification.

During image-guided surgery and anastomosis procedures, it is important to monitor physiological status and assess tissue perfusion. Hyperspectral imaging (HSI) can be used to estimate superficial tissue oxygen saturation (sO2), where native reflectance measurements are converted to absorbance for spectral unmixing and sO(2) estimation. However, keeping the reflectance data in its native form may improve accuracy. We developed a reflectance-based approach for sO(2) estimation with a high-speed HSI system, leveraging band-filtering and data processing techniques to improve performance. Our approach is validated with blood oxygenation and tissue-mimicking phantoms, utilizing a blood-gas analyzer (BGA) to obtain ground-truth measurements. The reflectance-based approach achieved a median error of 5.0% and 5.3% with the blood oxygenation and tissue-mimicking phantoms, respectively, effectively halving the error achieved with an absorbance-based approach (10.4%). Overall, we present a novel reflectance-based approach for sO2 estimation that achieves superior performance compared to a conventional absorbance-based technique. Paired with high-speed HSI and robust BGA validation, this approach offers strong potential for clinical translation. Thus, the reflectance-based HSI sO(2) estimation approach can have many applications in image-guided surgery and interventional procedures.

Purpose:During interventional procedures, clinicians need to mentally register anatomical information from preoperative cross-sectional images onto the patient's body to envision the location of subsurface targets and critical structures. This study aims to address this challenge by developing an augmented reality (AR)-based tracking system that can leverage cross-sectional images to provide real-time three-dimensional (3D) visualization of lesion targets and enable precise surgical procedures. Approach:Our AR platform combines a customized high-speed, real-time, optical tracking system, a holographic display device, a computer workstation, and a graphical user interface. The system displays holograms of virtual models of the target organ and surgical tools, as well as the navigation path. To validate our AR platform, we measured target registration errors (TRE) across two different types of interventional procedures using our high-speed optical tracking system. To conduct these experiments, we performed laparoscopic procedures and prostate biopsies on customized phantoms. Results:The integrated high-speed tracking and AR platform was applied to laparoscopic and biopsy procedures. The average TRE was 4.17 ± 1.63    mm for laparoscopic procedures and 2.89 ± 0.84    mm for prostate biopsy. Conclusions:An augmented reality platform with a high-speed precision optical tracking system was developed for interventional procedures. The AR platform has been demonstrated for potential applications in prostate laparoscopic and biopsy procedures and can be expanded to other interventional procedures.

Benign prostatic hyperplasia (BPH) is a common condition among aging men. Currently, transrectal ultrasonography (TRUS) is used for detecting BPH but frequently falls short in accuracy. In this study, we investigated the use of a deep learning network for stromal nodule detection in prostate specimens exhibiting BPH using B-mode ultrasound (US) imaging. B-mode US data was captured from ex- vivo prostate specimens of human patients. Expert annotations differentiating between stromal and non-stromal regions for each prostate specimen were correlated with corresponding B-mode data and used as the reference for validation. The Segment Anything Model 2 (SAM2) network was trained and applied to segment stromal versus non-stromal tissue within B-mode ultrasound images. An 80/20 patient-level train/test split was used to evaluate model performance. Preliminary results demonstrate the feasibility of using SAM2 to identify stromal nodules directly from B-mode ultrasound data, achieving a Dice score of 0.76. Ongoing work focuses on increasing the dataset and improving segmentation accuracy. The study provides an approach to validate the capability of ultrasound imaging for BPH detection. More image data are needed to improve the segmentation performance.

A deformable multi-organ abdominal phantom can have many applications in medical imaging and image-guided interventions. It can provide ground truth data for registration evaluation and procedure training. In this study, we designed and fabricated a deformable abdominal phantom for multimodal imaging applications. Computed tomography ( CT) images of a human patient were used to segment the abdominal surface, kidneys, spleen, and liver. Scaled down 3D-printed molds were created based on the segmented 3D models of these structures. To allow for deformability, an agar-gelatin gel mixture was used to cast the phantom using the 3D-printed molds with potassium iodide added to the target formulation as a CT contrast agent. Psyllium husk powder was utilized as an ultrasound (US) scattering agent. Simulated tumors or targets were incorporated into the kidney phantom. All components were assembled into an abdominal mold to replicate patient anatomical structures. The phantom was evaluated using CT and ultrasound imaging, with preliminary results demonstrating its ability to provide ground truth for tumor localization. This study presents a reproducible framework for creating a deformable abdominal phantom that can be applied to the development and validation of image-guided interventional procedures.

Minimally invasive procedures enable clinicians to perform interventions with reduced tissue damage and expedited recovery times. However, laparoscopic imaging inherently provides a limited field of view, compelling surgeons to mentally integrate preoperative imaging with intraoperative scenes, which is prone to localization errors. Augmented reality (AR) offers a solution by overlaying preoperative 3D organ models onto live laparoscopic video; however, precise 3D reconstruction and localization of the surgical scene is critical for accurate overlays. In this study, we evaluated three deep learning-based surface reconstruction methods, DROID-SLAM, MASt3R-SLAM, and the visual geometry grounded transformer (VGGT), for reconstructing tissue surfaces from laparoscopic video. We evaluated these algorithms with a dataset comprised of laparoscopic RGB frames of ex-vivo porcine tissue, ground-truth camera trajectories obtained via an optical tracking system and known camera intrinsic parameters. Dense point cloud and depth map predications were then qualitatively compared between models and estimated camera trajectory was compared quantitively using absolute trajectory error (ATE) and relative pose error (RPE). Preliminary results indicate that all models produced plausible tissue surface reconstructions, with estimated depth maps and camera trajectories demonstrating close alignment with the ground truth. These findings suggest that deep learning-based 3D reconstruction methods can effectively support image-guided interventions without relying on ground-truth camera poses or depth maps and may be integrated with AR to support minimally invasive surgical workflows.

Purpose:Identifying pregnant patients at high risk of hysterectomy before giving birth informs clinical management and improves outcomes. We aim to develop machine learning models to predict hysterectomy in pregnant women with placenta accreta spectrum (PAS). Approach:We developed five machine learning models using information from magnetic resonance images and combined them with topographic maps and radiomic features to predict hysterectomy. The models were trained, optimized, and evaluated on data from 241 patients, in groups of 157, 24, and 60 for training, validation, and testing, respectively. Results:We assessed the models individually as well as using an ensemble approach. When these models are combined, the ensembled model produced the best performance and achieved an area under the curve of 0.90, a sensitivity of 90.0%, and a specificity of 90.0% for predicting hysterectomy. Conclusions:Various machine learning models were developed to predict hysterectomy in pregnant women with PAS, which may have potential clinical applications to help improve patient management.

Isocitrate dehydrogenase (IDH) mutation status is a critical prognostic indicator in glioma patients. Numerous studies have focused on developing non-invasive methodologies to classify IDH status using pre-operative MRI scans. However, the challenge lies in data scarcity and class imbalance in IDH mutations. This study explores generative AI methods to augment training data and enhance IDH classification accuracy. We developed a 3D conditional latent diffusion model (LDM) for generating 3D multi-contrast brain tumor MRI data (128 × 128 × 64 with a voxel spacing of 1.5 × 1.5 × 2.0 mm) with whole tumor mask and IDH mutation status as conditions. The LDM comprises a 3D autoencoder for perceptual compression and a conditional 3D diffusion model (DM) for generating multi-contrast synthetic samples guided by tumor masks and the IDH mutation status. We incorporated two types of attention modules within the denoising UNet of the LDM to capture the semantic class-dependent data distribution driven by the provided whole tumor mask and IDH status. The LDM was trained using two brain tumor datasets: The Cancer Genome Atlas dataset and an internal dataset from the University of Texas Southwestern Medical Center. The synthetic images generated by the LDM were then used to train IDH classification models, which were subsequently tested on real brain tumor data comprising 327 mutated and 1,394 wild-type cases from the University of California San Francisco Preoperative Diffuse Glioma MRI dataset, the Erasmus Glioma Database, the University of Pennsylvania glioblastoma, and two held-out internal datasets. The IDH classification models, trained on synthetic images and tested on real data, achieved an excellent overall classification accuracy of 94.02%. This approach has the potential to be extended to other molecular markers where data scarcity presents a challenge.