Mohammed Baharoon

Jun Ma, Sumin Kim, Feifei Li et al.

Recent advances in segmentation foundation models have enabled accurate and efficient segmentation across a wide range of natural images and videos, but their utility to medical data remains unclear. In this work, we first present a comprehensive benchmarking of the Segment Anything Model 2 (SAM2) across 11 medical image modalities and videos and point out its strengths and weaknesses by comparing it to SAM1 and MedSAM. Then, we develop a transfer learning pipeline and demonstrate SAM2 can be quickly adapted to medical domain by fine-tuning. Furthermore, we implement SAM2 as a 3D slicer plugin and Gradio API for efficient 3D image and video segmentation. The code has been made publicly available at \url{https://github.com/bowang-lab/MedSAM}.

Mohammed Baharoon, Thibault Heintz, Siavash Raissi et al.

We introduce CRIMSON, a clinically grounded evaluation framework for chest X-ray report generation that assesses reports based on diagnostic correctness, contextual relevance, and patient safety. Unlike prior metrics, CRIMSON incorporates full clinical context, including patient age, indication, and guideline-based decision rules, and prevents normal or clinically insignificant findings from exerting disproportionate influence on the overall score. The framework categorizes errors into a comprehensive taxonomy covering false findings, missing findings, and eight attribute-level errors (e.g., location, severity, measurement, and diagnostic overinterpretation). Each finding is assigned a clinical significance level (urgent, actionable non-urgent, non-actionable, or expected/benign), based on a guideline developed in collaboration with attending cardiothoracic radiologists, enabling severity-aware weighting that prioritizes clinically consequential mistakes over benign discrepancies. CRIMSON is validated through strong alignment with clinically significant error counts annotated by six board-certified radiologists in ReXVal (Kendalls tau = 0.61-0.71; Pearsons r = 0.71-0.84), and through two additional benchmarks that we introduce. In RadJudge, a targeted suite of clinically challenging pass-fail scenarios, CRIMSON shows consistent agreement with expert judgment. In RadPref, a larger radiologist preference benchmark of over 100 pairwise cases with structured error categorization, severity modeling, and 1-5 overall quality ratings from three cardiothoracic radiologists, CRIMSON achieves the strongest alignment with radiologist preferences. We release the metric, the evaluation benchmarks, RadJudge and RadPref, and a fine-tuned MedGemma model to enable reproducible evaluation of report generation, all available at https://github.com/rajpurkarlab/CRIMSON.

Mohammed Baharoon, Hessa Almatar, Reema Alduhayan et al.

In recent years, deep learning has shown promise in predicting hypertension (HTN) from fundus images. However, most prior research has primarily focused on analyzing a single type of data, which may not capture the full complexity of HTN risk. To address this limitation, this study introduces a multimodal deep learning (MMDL) system, dubbed HyMNet, which combines fundus images and cardiometabolic risk factors, specifically age and gender, to improve hypertension detection capabilities. Our MMDL system uses RETFound, a foundation model pre-trained on 1.6 million retinal images, for the fundus path and a fully connected neural network for the age and gender path. The two paths are jointly trained by concatenating the feature vectors from each path that are then fed into a fusion network. The system was trained on 5,016 retinal images from 1,243 individuals collected from the Saudi Ministry of National Guard Health Affairs. The results show that the multimodal model that integrates fundus images along with age and gender outperforms the unimodal system trained solely on fundus photographs, with an F1 score of 0.771 [0.747, 0.796], and 0.745 [0.719, 0.772] for hypertension detection, respectively. Additionally, we studied the effect underlying diabetes mellitus has on the model's predictive ability, concluding that diabetes is used as a confounding variable for distinguishing hypertensive cases. Our code and model weights are publicly available at https://github.com/MohammedSB/HyMNet.

Oishi Banerjee, Sung Eun Kim, Alexandra N. Willauer et al.

Everyday photographs taken with ordinary cameras are already widely used in telemedicine and other online health conversations, yet no comprehensive benchmark evaluates whether vision-language models can interpret their medical content. Analyzing these images requires both fine-grained natural image understanding and domain-specific medical reasoning, a combination that challenges both general-purpose and specialized models. We introduce ReXInTheWild, a benchmark of 955 clinician-verified multiple-choice questions spanning seven clinical topics across 484 photographs sourced from the biomedical literature. When evaluated on ReXInTheWild, leading multimodal large language models show substantial performance variation: Gemini-3 achieves 78% accuracy, followed by Claude Opus 4.5 (72%) and GPT-5 (68%), while the medical specialist model MedGemma reaches only 37%. A systematic error analysis also reveals four categories of common errors, ranging from low-level geometric errors to high-level reasoning failures and requiring different mitigation strategies. ReXInTheWild provides a challenging, clinically grounded benchmark at the intersection of natural image understanding and medical reasoning. The dataset is available on HuggingFace.

Mohammed Baharoon, Waseem Qureshi, Jiahong Ouyang et al.

The integration of deep learning systems into healthcare has been hindered by the resource-intensive process of data annotation and the inability of these systems to generalize to different data distributions. Foundation models, which are models pre-trained on large datasets, have emerged as a solution to reduce reliance on annotated data and enhance model generalizability and robustness. DINOv2 is an open-source foundation model pre-trained with self-supervised learning on 142 million curated natural images that exhibits promising capabilities across various vision tasks. Nevertheless, a critical question remains unanswered regarding DINOv2's adaptability to radiological imaging, and whether its features are sufficiently general to benefit radiology image analysis. Therefore, this study comprehensively evaluates the performance DINOv2 for radiology, conducting over 200 evaluations across diverse modalities (X-ray, CT, and MRI). To measure the effectiveness and generalizability of DINOv2's feature representations, we analyze the model across medical image analysis tasks including disease classification and organ segmentation on both 2D and 3D images, and under different settings like kNN, few-shot learning, linear-probing, end-to-end fine-tuning, and parameter-efficient fine-tuning. Comparative analyses with established supervised, self-supervised, and weakly-supervised models reveal DINOv2's superior performance and cross-task generalizability. The findings contribute insights to potential avenues for optimizing pre-training strategies for medical imaging and enhancing the broader understanding of DINOv2's role in bridging the gap between natural and radiological image analysis. Our code is available at https://github.com/MohammedSB/DINOv2ForRadiology

Mohammed Baharoon, Luyang Luo, Michael Moritz et al.

We introduce ReXGroundingCT, the first publicly available dataset linking free-text findings to pixel-level 3D segmentations in chest CT scans. The dataset includes 3,142 non-contrast chest CT scans paired with standardized radiology reports from CT-RATE. Construction followed a structured three-stage pipeline. First, GPT-4 was used to extract and standardize findings, descriptors, and metadata from reports originally written in Turkish and machine-translated into English. Second, GPT-4o-mini categorized each finding into a hierarchical ontology of lung and pleural abnormalities. Third, 3D annotations were produced for all CT volumes: the training set was quality-assured by board-certified radiologists, and the validation and test sets were fully annotated by board-certified radiologists. Additionally, a complementary chain-of-thought dataset was created to provide step-by-step hierarchical anatomical reasoning for localizing findings within the CT volume, using GPT-4o and localization coordinates derived from organ segmentation models. ReXGroundingCT contains 16,301 annotated entities across 8,028 text-to-3D-segmentation pairs, covering diverse radiological patterns from 3,142 non-contrast CT scans. About 79% of findings are focal abnormalities and 21% are non-focal. The dataset includes a public validation set of 50 cases and a private test set of 100 cases, both annotated by board-certified radiologists. The dataset establishes a foundation for enabling free-text finding segmentation and grounded radiology report generation in CT imaging. Model performance on the private test set is hosted on a public leaderboard at https://rexrank.ai/ReXGroundingCT. The dataset is available at https://huggingface.co/datasets/rajpurkarlab/ReXGroundingCT.

Sraavya Sambara, Sung Eun Kim, Xiaoman Zhang et al.

Current Vision-Language Models (VLMs) struggle to ground anatomical regions in 3D medical images and reason about them in a step-by-step manner, a key requirement of real-world diagnostic assessment. This ability is essential for aligning model outputs with the diagnostic workflows clinicians use in practice, enabling trustworthy clinician-AI collaboration. Existing 3D datasets provide localization labels, but none support this "grounded reasoning" ability. To address this gap, we introduce 3DReasonKnee, the first 3D grounded reasoning dataset for medical images, which provides 494k high-quality quintuples derived from 7,970 3D knee MRI volumes. Each quintuple includes: (1) the 3D MRI volume, (2) a diagnostic question targeting a specific anatomical region (3) a 3D bounding box localizing the relevant anatomical structures, (4) clinician-generated diagnostic reasoning steps that explicitly detail the 3D reasoning process, and (5) structured severity assessments for the relevant anatomical region. The creation and validation of 3DReasonKnee, involving over 450 hours of expert clinician time for manually segmenting MRIs and generating reasoning chains, ensures its superior quality and clinical relevance. We establish ReasonKnee-Bench to evaluate localization and diagnostic accuracy, providing insight into VLM ability to perform grounding and severity assessment across anatomical regions and diagnostic inquiries. We benchmark five state-of-the-art VLMs, providing baseline performance for ReasonKnee-Bench. By providing this unique resource of expert-annotated 3D reasoning pathways, 3DReasonKnee serves as a repository of orthopedic surgeons' diagnostic expertise and offers a vital testbed for advancing multimodal medical AI systems towards 3D, clinically aligned, localized decision-making capabilities. The dataset can be found in: https://huggingface.co/datasets/rajpurkarlab/3DReasonKnee

Mohammed Baharoon, Jun Ma, Congyu Fang et al.

Multimodal Large Language Models (MLLMs) have emerged as a promising way to automate Radiology Report Generation (RRG). In this work, we systematically investigate the design space of 3D MLLMs, including visual input representation, projectors, Large Language Models (LLMs), and fine-tuning techniques for 3D CT report generation. We also introduce two knowledge-based report augmentation methods that improve performance on the GREEN score by up to 10%, achieving the 2nd place on the MICCAI 2024 AMOS-MM challenge. Our results on the 1,687 cases from the AMOS-MM dataset show that RRG is largely independent of the size of LLM under the same training protocol. We also show that larger volume size does not always improve performance if the original ViT was pre-trained on a smaller volume size. Lastly, we show that using a segmentation mask along with the CT volume improves performance. The code is publicly available at https://github.com/bowang-lab/AMOS-MM-Solution

Jun Ma, Zongxin Yang, Sumin Kim et al.

Medical image and video segmentation is a critical task for precision medicine, which has witnessed considerable progress in developing task or modality-specific and generalist models for 2D images. However, there have been limited studies on building general-purpose models for 3D images and videos with comprehensive user studies. Here, we present MedSAM2, a promptable segmentation foundation model for 3D image and video segmentation. The model is developed by fine-tuning the Segment Anything Model 2 on a large medical dataset with over 455,000 3D image-mask pairs and 76,000 frames, outperforming previous models across a wide range of organs, lesions, and imaging modalities. Furthermore, we implement a human-in-the-loop pipeline to facilitate the creation of large-scale datasets resulting in, to the best of our knowledge, the most extensive user study to date, involving the annotation of 5,000 CT lesions, 3,984 liver MRI lesions, and 251,550 echocardiogram video frames, demonstrating that MedSAM2 can reduce manual costs by more than 85%. MedSAM2 is also integrated into widely used platforms with user-friendly interfaces for local and cloud deployment, making it a practical tool for supporting efficient, scalable, and high-quality segmentation in both research and healthcare environments.

Mohammed Baharoon, Siavash Raissi, John S. Jun et al.

We introduce RadGame, an AI-powered gamified platform for radiology education that targets two core skills: localizing findings and generating reports. Traditional radiology training is based on passive exposure to cases or active practice with real-time input from supervising radiologists, limiting opportunities for immediate and scalable feedback. RadGame addresses this gap by combining gamification with large-scale public datasets and automated, AI-driven feedback that provides clear, structured guidance to human learners. In RadGame Localize, players draw bounding boxes around abnormalities, which are automatically compared to radiologist-drawn annotations from public datasets, and visual explanations are generated by vision-language models for user missed findings. In RadGame Report, players compose findings given a chest X-ray, patient age and indication, and receive structured AI feedback based on radiology report generation metrics, highlighting errors and omissions compared to a radiologist's written ground truth report from public datasets, producing a final performance and style score. In a prospective evaluation, participants using RadGame achieved a 68% improvement in localization accuracy compared to 17% with traditional passive methods and a 31% improvement in report-writing accuracy compared to 4% with traditional methods after seeing the same cases. RadGame highlights the potential of AI-driven gamification to deliver scalable, feedback-rich radiology training and reimagines the application of medical AI resources in education.

Julian Acosta, Scott Adams, Julius Kernbach et al.

We developed a voice-driven artificial intelligence (AI) system that guides anyone - from paramedics to family members - through expert-level stroke evaluations using natural conversation, while also enabling smartphone video capture of key examination components for documentation and potential expert review. This addresses a critical gap in emergency care: current stroke recognition by first responders is inconsistent and often inaccurate, with sensitivity for stroke detection as low as 58%, causing life-threatening delays in treatment. Three non-medical volunteers used our AI system to assess ten simulated stroke patients, including cases with likely large vessel occlusion (LVO) strokes and stroke-like conditions, while we measured diagnostic accuracy, completion times, user confidence, and expert physician review of the AI-generated reports. The AI system correctly identified 84% of individual stroke signs and detected 75% of likely LVOs, completing evaluations in just over 6 minutes. Users reported high confidence (median 4.5/5) and ease of use (mean 4.67/5). The system successfully identified 86% of actual strokes but also incorrectly flagged 2 of 3 non-stroke cases as strokes. When an expert physician reviewed the AI reports with videos, they identified the correct diagnosis in 100% of cases, but felt confident enough to make preliminary treatment decisions in only 40% of cases due to observed AI errors including incorrect scoring and false information. While the current system's limitations necessitate human oversight, ongoing rapid advancements in speech-to-speech AI models suggest that future versions are poised to enable highly accurate assessments. Achieving human-level voice interaction could transform emergency medical care, putting expert-informed assessment capabilities in everyone's hands.

Mohammed Baharoon, Jonathan Klein, Dominik L. Michels

Vision-language contrastive learning frameworks such as CLIP enable learning representations from natural language supervision and provide strong zero-shot classification capabilities. However, due to the nature of the supervisory signal in these paradigms, they lack the ability to learn localized features, leading to degraded performance on dense prediction tasks such as segmentation and detection. On the other hand, self-supervised learning methods have shown the ability to learn granular representations, complementing the high-level features in vision-language training. In this work, we present Harmony, a framework that combines vision-language training with discriminative and generative self-supervision to learn visual features that can be generalized across different downstream vision tasks. Our framework is specifically designed to work on web-scraped data by not relying on negative examples in the self-supervised learning path and addressing the one-to-one correspondence issue using soft CLIP targets generated by an EMA model. Moreover, Harmony optimizes for five different objectives simultaneously, efficiently utilizing the supervision in each data example, making it even more suited in data-constrained settings. We comprehensively evaluate Harmony across various vision downstream tasks and find that it significantly outperforms the baseline CLIP and outperforms the previously leading joint self- and weakly supervised methods, SLIP, MaskCLIP, and DetailCLIP.