Optical coherence tomography (OCT) images are often acquired as highly anisotropic volumes, where the scanning step is dense along the fast axis but sparse along the slow axis. This affects image analysis, such as image registration for longitudinal alignment. To create more isotropic volumes, bicubic interpolation can be used along the slow axis, but it generally produces blurry features. Registration-based interpolation can reduce blurriness, but often fails to generate realistic OCT images. Deep generative models can sample realistic images, but lack the structural consistency constraints required for interpolation. In this paper, we propose an unsupervised image interpolation method that combines registration-based interpolation with a deep generative model to overcome their individual limitations and improve the structural accuracy and realism of interpolated OCT images. We compare the proposed method with both bicubic and registration-based interpolation on real OCT datasets, and show that it achieves the best interpolation performance.

Accurately predicting visual field progression is critical for early intervention and personalized treatment of glaucoma. However, existing methods struggle with both predictive accuracy and reliable uncertainty quantification. This paper introduces a framework that leverages diffusion models and conformal risk control to generate robust and interpretable forecasts of visual field deterioration. We first train a diffusion model to predict future visual fields based on a patient's past examinations. To ensure trustworthy predictions, we design a novel archetypal-based conformal risk control method, which provides finite-sample coverage guarantees on intervals of archetypal contributions. This framework captures the underlying structures within uncertainty, enabling clinicians to interpret a range of potential progression patterns rather than a single deterministic outcome. Experimental results illustrate that our method achieves the target archetypal contribution coverage while providing tighter prediction intervals than baselines. Visualizations show how archetypal visual field patterns contribute to prediction uncertainty, offering interpretable insights into disease progression. By combining diffusion models with conformal methods, our framework enhances the reliability of AI-assisted visual field forecasting, ultimately supporting improved clinical decision-making. Our code is available at: https://github.com/averysi224/abci.git.

Fluorescence lifetime (FLT) imaging has been shown to distinguish tumors from normal tissue with high accuracy. However, the practical utility of FLT imaging is hindered by slow acquisition speeds and depth-dependent inaccuracies. To address these challenges, we introduce FLT-SLAM, a novel algorithm that combines rapid FLT imaging with simultaneous localization and mapping (SLAM) for real-time 3D surface reconstruction and depth-corrected FLT estimation. Using a stereo laparoscope, our approach extracts real-time depth information to improve accuracy, while achieving acquisition speeds exceeding 5 Hz. FLT maps are overlaid onto large-scale 3D surface models generated by SLAM, improving visualization and spatial awareness. We validate FLT-SLAM through phantom and ex-vivo tissue measurements, and show that it reduces FLT estimation errors by nearly 20%, thereby demonstrating its potential to enhance real-time, depth-corrected FLT imaging for surgical applications.

Reliable 3D reconstruction of tissue architecture from sequential 2D multiplex images is challenging due to the noise and distortions introduced by ultrathin (50 nm) slicing and complex alignment procedures. Conventional cell tracking methods often fail under such conditions, resulting in inaccurate linkage of cells across sections. To bridge this gap, we propose a Bayesian Transformer framework that incorporates uncertainty-aware feature embeddings and higher-order graph matching with belief propagation. By tracking cells across consecutive sections, our method facilitates the 3D reconstruction of volumetric tissue organization, even in highly noise-prone scenarios. The methodology begins with a standard segmentation step, followed by feature extraction that computes morphological, shape, and texture descriptors, as well as deep CNN embeddings. These rich, uncertainty-sensitive representations reduce errors caused by both registration artifacts and morphological variability. We validate the effectiveness of the proposed approach on a private multiplex dataset of fixed tissue sections and further demonstrate its generalizability on public time-lapse microscopy videos, showcasing adaptability to diverse datasets. Experimental comparisons reveal that our method outperforms baseline tracking techniques, achieving higher accuracy and more consistent cell linkages across multiple serial sections. The code used in this research with sample dataset are publicly available at https://github.com/NabaviLab/bayesian-transformer-cell-tracking.

Mammography screening is an essential tool for early detection of breast cancer. The speed and accuracy of mammography interpretation has the potential to be improved with deep learning methods. However, the development of a foundation visual language model (VLM) is hindered by limited data and domain differences between natural and medical images. Existing mammography VLMs, adapted from natural images, often ignore domain-specific characteristics, such as multi-view relationships in mammography. Unlike radiologists who analyze both views together to process ipsilateral correspondence, current methods treat them as independent images or do not properly model the multi-view correspondence learning, losing critical geometric context and resulting in suboptimal prediction. We propose GLAM: Global and Local Alignment for Multi-view mammography for VLM pretraining using geometry guidance. By leveraging the prior knowledge about the multi-view imaging process of mammograms, our model learns local cross-view alignments and fine-grained local features through joint global and local, visual-visual, and visual-language contrastive learning. Pretrained on EMBED [14], one of the largest open mammography datasets, our model outperforms baselines across multiple datasets under different settings.

Generative models enhance neuroimaging through data augmentation, quality improvement, and rare condition studies. Despite advances in realistic synthetic MRIs, evaluations focus on texture and perception, lacking sensitivity to crucial morphometric fidelity. This study proposes a new metric, called WASABI (Wasserstein-Based Anatomical Brain Index), to assess the morphometric plausibility of synthetic brain MRIs. WASABI leverages SynthSeg, a deep learning-based brain parcellation tool, to derive volumetric measures of brain regions in each MRI and uses the multivariate Wasserstein distance to compare distributions between real and synthetic anatomies. Based on controlled experiments on two real datasets and synthetic MRIs from five generative models, WASABI demonstrates higher sensitivity in quantifying morphometric discrepancies compared to traditional image-level metrics, even when synthetic images achieve near-perfect visual quality. Our findings advocate for shifting the evaluation paradigm beyond visual inspection and conventional metrics, emphasizing morphometric fidelity as a crucial benchmark for clinically meaningful brain MRI synthesis. Our code is available at https://github.com/BahramJafrasteh/wasabi-mri.

While imaging-genetics holds great promise for unraveling the complex interplay between brain structure and genetic variation in neurological disorders, traditional methods are limited to simplistic linear models or to black-box techniques that lack interpretability. In this paper, we present NeuroPathX, an explainable deep learning framework that uses an early fusion strategy powered by cross-attention mechanisms to capture meaningful interactions between structural variations in the brain derived from MRI and established biological pathways derived from genetics data. To enhance interpretability and robustness, we introduce two loss functions over the attention matrix - a sparsity loss that focuses on the most salient interactions and a pathway similarity loss that enforces consistent representations across the cohort. We validate NeuroPathX on both autism spectrum disorder and Alzheimer's disease. Our results demonstrate that NeuroPathX outperforms competing baseline approaches and reveals biologically plausible associations linked to the disorder. These findings underscore the potential of NeuroPathX to advance our understanding of complex brain disorders.

We propose a graph information compression framework, called Behavior-Informed Subgroup-consistent Connectome Template (BISCoT), that learns interpretable functional subnetworks from resting-state fMRI (rs-fMRI) connectivity, which simultaneously capture the heterogeneity of a diverse patient cohort. BISCoT uses multidimensional behavioral profiles to guide the decomposition of a rs-fMRI connectivity matrices into sparse yet representative subnetworks that are consistent within behavioral sub-groups. In particular, our framework adopts a graph convolution network to capture local connectivity features and applies a subgroup-informed pooling process to extract the final subnetworks. We evaluate BISCoT on an in-house dataset of individuals with autism spectrum disorder and demonstrate that the learned subnetworks improve the performance of multiple downstream prediction tasks. In addition, BISCoT extracts consistent connectivity "templates" at the subgroup level, which allows for interpretable biomarker identification.

While artificial intelligence (AI) has revolutionized the field of epileptic seizure detection from electroencephalography (EEG), its clinical adoption remains limited, largely due to the lack of transparency in AI models and their inability to explain the underlying seizure etiology. This paper introduces SzXAI, a novel framework to enhance the reasoning abilities of AI models for EEG-based seizure detection. SzXAI employs a contrastive training mechanism, which uses cross-modality similarity layers to align the EEG encodings with textual concept embeddings derived from clinical notes using LLMs. Along with the alignment, SzXAI leverages an attention-weighted pooling mechanism to detect underlying seizure and baseline etiologies. We validate SzXAI via 10-fold cross validation on the publicly available Temple University Hospital dataset. Our results demonstrate that the alignment-powered training mechanism of SzXAI vastly outperforms direct etiology prediction, thus improving the reliability of the predicted seizure etiologies. Furthermore, structured sentence generation using the model output provided insights in a human-readable format. Thus, SzXAI provides an effective platform to boost clinical trust and AI usability in epilepsy management.

We introduce a novel composite total variation (TV) and its solution algorithm with their application to multi-echo, respiratory motion-resolved 5D (3D space + 1D respiratory motion + 1D echo signal evolution) compressed sensing (CS) abdominal MR image reconstruction. The proposed formalism ensures a sparse representation between multi-echo images with varying contrast-a vital feature that needs to be preserved-making it highly suitable for applications in multi-dimensional computational/quantitative imaging. The key idea of the proposed composite TV and its formal definition were inspired by the observation that the spatial gradient of difference images in multi-echo MRI appears sparse. Throughout extensive experiments on a small number of healthy volunteers, we have demonstrated improved performance of the proposed method in 5D motion-resolved CS reconstruction of multi-echo MRI data compared to the state-of-the-art method. We have also demonstrated improved performance of the proposed method in quantitative tissue parameter mapping (such as R2*, proton density fat fraction, and quantitative susceptibility mapping) across a wide range of undersampling factors. In conclusion, the proposed method enables vastly accelerated motion-resolved multi-echo CS-MRI minimally impacting the quantification of downstream tissue parameters.

Analyzing fetal body motion and shape is paramount in prenatal diagnostics and monitoring. Existing methods for fetal MRI analysis mainly rely on anatomical keypoints or volumetric body segmentations. Keypoints simplify body structure to facilitate motion analysis, but may ignore important details of full-body shape. Body segmentations capture complete shape information but complicate temporal analysis due to large non-local fetal movements. To address these limitations, we construct a 3D articulated statistical fetal body model based on the Skinned Multi-Person Linear Model (SMPL). Our algorithm iteratively estimates body pose in the image space and body shape in the canonical pose space. This approach improves robustness to MRI motion artifacts and intensity distortions, and reduces the impact of incomplete surface observations due to challenging fetal poses. We train our model on segmentations and keypoints derived from 19, 816 MRI volumes across 53 subjects. Our model captures body shape and motion across time series and provides intuitive visualization. Furthermore, it enables automated anthropometric measurements traditionally difficult to obtain from segmentations and keypoints. When tested on unseen fetal body shapes, our method yields a surface alignment error of 3.2 mm for 3 mm MRI voxel size. To our knowledge, this represents the first 3D articulated statistical fetal body model, paving the way for enhanced fetal motion and shape analysis in prenatal diagnostics. The code is available at https://github.com/MedicalVisionGroup/fetal-smpl.

Fetal motion is a critical indicator of neurological development and intrauterine health, yet its quantification remains challenging, particularly at earlier gestational ages (GA). Current methods track fetal motion by predicting the location of annotated landmarks on 3D echo planar imaging (EPI) time-series, primarily in third-trimester fetuses. The predicted landmarks enable simplification of the fetal body for downstream analysis. While these methods perform well within their training age distribution, they consistently fail to generalize to early GAs due to significant anatomical changes in both mother and fetus across gestation, as well as the difficulty of obtaining annotated early GA EPI data. In this work, we develop a cross-population data augmentation framework that enables pose estimation models to robustly generalize to younger GA clinical cohorts using only annotated images from older GA cohorts. Specifically, we introduce a fetal-specific augmentation strategy that simulates the distinct intrauterine environment and fetal positioning of early GAs. Our experiments find that cross-population augmentation yields reduced variability and significant improvements across both older GA and challenging early GA cases. By enabling more reliable pose estimation across gestation, our work potentially facilitates early clinical detection and intervention in challenging 4D fetal imaging settings. Code is available at https://github.com/sebodiaz/cross-population-pose.

Vision-Language Models (VLMs) are becoming increasingly popular in the medical domain, bridging the gap between medical images and clinical language. Existing VLMs demonstrate an impressive ability to comprehend medical images and text queries to generate detailed, descriptive diagnostic medical reports. However, hallucination-the tendency to generate descriptions that are inconsistent with the visual content-remains a significant issue in VLMs, with particularly severe implications in the medical field. To facilitate VLM research on gastrointestinal (GI) image analysis and study hallucination, we curate a multimodal image-text GI dataset: Gut-VLM. This dataset is created using a two-stage pipeline: first, descriptive medical reports of Kvasir-v2 images are generated using ChatGPT, which introduces some hallucinated or incorrect texts. In the second stage, medical experts systematically review these reports, and identify and correct potential inaccuracies to ensure high-quality, clinically reliable annotations. Unlike traditional datasets that contain only descriptive texts, our dataset also features tags identifying hallucinated sentences and their corresponding corrections. A common approach to reducing hallucination in VLM is to finetune the model on a small-scale, problem-specific dataset. However, we take a different strategy using our dataset. Instead of finetuning the VLM solely for generating textual reports, we finetune it to detect and correct hallucinations, an approach we call hallucination-aware finetuning. Our results show that this approach is better than simply finetuning for descriptive report generation. Additionally, we conduct an extensive evaluation of state-of-the-art VLMs across several metrics, establishing a benchmark.

The development of a computational framework that can infer large-scale brain-wide effective connectivity (EC) based on resting-state functional MRI (rs-fMRI) represents a grand challenge to computational neuroimaging. Towards the goal of estimating full-scale, whole-brain EC, we developed a new computational framework termed Large-scale nEural Model Inversion (LEMI) by utilizing a linear neural mass model with an efficient Kalman-filter based gradient descent algorithm. Key advantages of LEMI include fast estimation of both intra-regional and inter-regional connection strengths for large-scale networks, allowing exploration of both intrinsic and external mechanisms in neuroscience problems. Using ground-truth simulations, we demonstrated that LEMI can accurately and efficiently recover model parameters in a large network (100 regions) within 90 minutes. We then applied the LEMI model to an empirical rs-fMRI dataset from the ADNI database and identified widespread reduced excitation-inhibition (E-I) ratio in patients with Alzheimer's disease (AD). Overall, LEMI provides an efficient and accurate computational framework to estimate large-scale EC and whole-brain E-I balance based on non-invasive neuroimaging data.

Tau pathology is a hallmark of Alzheimer's disease (AD), and longitudinal tau positron emission tomography (PET) provides valuable insights into disease progression. However, the integration of tau PET data into computational models remains limited by challenges in encoding topographical information and ensuring longitudinal consistency. Existing biomarker-based representations often lack spatial flexibility and fail to account for covariance between brain regions. Additionally, traditional approaches often treat longitudinal scans as independent observations, neglecting temporal coherence. To address these limitations, we propose a novel Multiresolutional Reeb Graph representation that encodes the spatiotemporal propagation of tau topographical information. Our method constructs Reeb graphs to capture tau topography at a static time point and extends them into a multiresolutional framework to model disease evolution. We introduce a topology-based measurement for quantifying pathology spatial distribution similarity, and a severity interleaving distance for robust longitudinal staging. The efficiency of the proposed representation is validated in two downstream tasks: an integrated subtyping and staging system, and the longitudinal pathology prediction. The promising results compared with the current methods demonstrate the great potential of the proposed representation to enhancing the application of longitudinal tau PET data, and offering a reliable approach for studying AD progression.