Exploring the trustworthiness of deep learning models is crucial, especially in critical domains such as medical imaging decision support systems. Conformal prediction has emerged as a rigorous means of providing deep learning models with reliable uncertainty estimates and safety guarantees. However, conformal prediction results face challenges due to the backbone model's struggles in domain-shifted scenarios, such as variations in different sources. To aim this challenge, this paper proposes a novel framework termed Conformal Ensemble of Vision Transformers (CE-ViTs) designed to enhance image classification performance by prioritizing domain adaptation and model robustness, while accounting for uncertainty. The proposed method leverages an ensemble of vision transformer models in the backbone, trained on diverse datasets including HAM10000, Dermofit, and Skin Cancer ISIC datasets. This ensemble learning approach, calibrated through the combined mentioned datasets, aims to enhance domain adaptation through conformal learning. Experimental results underscore that the framework achieves a high coverage rate of 90.38\%, representing an improvement of 9.95\% compared to the HAM10000 model. This indicates a strong likelihood that the prediction set includes the true label compared to singular models. Ensemble learning in CE-ViTs significantly improves conformal prediction performance, increasing the average prediction set size for challenging misclassified samples from 1.86 to 3.075.

Foundation models leveraging vision-language pretraining have shown promise in chest X-ray (CXR) interpretation, yet their real-world performance across diverse populations and diagnostic tasks remains insufficiently evaluated. This study benchmarks the diagnostic performance and generalizability of foundation models versus traditional convolutional neural networks (CNNs) on multinational CXR datasets. We evaluated eight CXR diagnostic models - five vision-language foundation models and three CNN-based architectures - across 37 standardized classification tasks using six public datasets from the USA, Spain, India, and Vietnam, and three private datasets from hospitals in China. Performance was assessed using AUROC, AUPRC, and other metrics across both shared and dataset-specific tasks. Foundation models outperformed CNNs in both accuracy and task coverage. MAVL, a model incorporating knowledge-enhanced prompts and structured supervision, achieved the highest performance on public (mean AUROC: 0.82; AUPRC: 0.32) and private (mean AUROC: 0.95; AUPRC: 0.89) datasets, ranking first in 14 of 37 public and 3 of 4 private tasks. All models showed reduced performance on pediatric cases, with average AUROC dropping from 0.88 +/- 0.18 in adults to 0.57 +/- 0.29 in children (p = 0.0202). These findings highlight the value of structured supervision and prompt design in radiologic AI and suggest future directions including geographic expansion and ensemble modeling for clinical deployment. Code for all evaluated models is available at https://drive.google.com/drive/folders/1B99yMQm7bB4h1sVMIBja0RfUu8gLktCE

Medical image segmentation is critical for disease diagnosis, treatment planning, and prognosis assessment, yet the complexity and diversity of medical images pose significant challenges to accurate segmentation. While Convolutional Neural Networks capture local features and Vision Transformers excel in the global context, both struggle with efficient long-range dependency modeling. Inspired by Mamba's State Space Modeling efficiency, we propose ÆMMamba, a novel multi-scale feature extraction framework built on the Mamba backbone network. AÆMMamba integrates several innovative modules: the Efficient Fusion Bridge (EFB) module, which employs a bidirectional state-space model and attention mechanisms to fuse multi-scale features; the Edge-Aware Module (EAM), which enhances low-level edge representation using Sobel-based edge extraction; and the Boundary Sensitive Decoder (BSD), which leverages inverse attention and residual convolutional layers to handle cross-level complex boundaries. ÆMMamba achieves state-of-the-art performance across 8 medical segmentation datasets. On polyp segmentation datasets (Kvasir, ClinicDB, ColonDB, EndoScene, ETIS), it records the highest mDice and mIoU scores, outperforming methods like MADGNet and Swin-UMamba, with a standout mDice of 72.22 on ETIS, the most challenging dataset in this domain. For lung and breast segmentation, ÆMMamba surpasses competitors such as H2Former and SwinUnet, achieving Dice scores of 84.24 on BUSI and 79.83 on COVID-19 Lung. And on the LGG brain MRI dataset, ÆMMamba attains an mDice of 87.25 and an mIoU of 79.31, outperforming all compared methods. The source code will be released at https://github.com/xingbod/eMMamba.

<i>"Just Accepted" papers have undergone full peer review and have been accepted for publication in <i>Radiology: Artificial Intelligence</i>. This article will undergo copyediting, layout, and proof review before it is published in its final version. Please note that during production of the final copyedited article, errors may be discovered which could affect the content.</i> Purpose To develop a deep learning segmentation model that can segment abdominal organs on CT and MR images with high accuracy and generalization ability. Materials and Methods In this study, an extended nnU-Net model was trained for abdominal organ segmentation. A domain randomization method in both the image and feature space was developed to improve the generalization ability under cross-site and cross-modality settings on public prostate MRI and abdominal CT and MRI datasets. The prostate MRI dataset contains data from multiple health care institutions with domain shifts. The abdominal CT and MRI dataset is structured for cross-modality evaluation, training on one modality (eg, MRI) and testing on the other (eg, CT). This domain randomization method was then used to train a segmentation model with enhanced generalization ability on the abdominal multiorgan segmentation challenge (AMOS) dataset to improve abdominal CT and MR multiorgan segmentation, and the model was compared with two commonly used segmentation algorithms (TotalSegmentator and MRSegmentator). Model performance was evaluated using the Dice similarity coefficient (DSC). Results The proposed domain randomization method showed improved generalization ability on the cross-site and cross-modality datasets compared with the state-of-the-art methods. The segmentation model using this method outperformed two other publicly available segmentation models on data from unseen test domains (Average DSC: 0.88 versus 0.79; <i>P</i> < .001 and 0.88 versus 0.76; <i>P</i> < .001). Conclusion The combination of image and feature domain randomizations improved the accuracy and generalization ability of deep learning-based abdominal segmentation on CT and MR images. © RSNA, 2025.

This study systematically examines the impact of training database size and the generalizability of deep learning models for synthetic medical image generation. Specifically, we employ a Cycle-Consistency Generative Adversarial Network (CycleGAN) with softly paired data to synthesize kilovoltage computed tomography (kVCT) images from megavoltage computed tomography (MVCT) scans. Unlike previous works, which were constrained by limited data availability, our study uses an extensive database comprising 4,000 patient CT scans, an order of magnitude larger than prior research, allowing for a more rigorous assessment of database size in medical image translation. We quantitatively evaluate the fidelity of the generated synthetic images using established image similarity metrics, including Mean Absolute Error (MAE) and Structural Similarity Index Measure (SSIM). Beyond assessing image quality, we investigate the model's capacity for generalization by analyzing its performance across diverse patient subgroups, considering factors such as sex, age, and anatomical region. This approach enables a more granular understanding of how dataset composition influences model robustness.

Self-supervised learning has emerged as a powerful paradigm for training deep neural networks, particularly in medical imaging where labeled data is scarce. While current approaches typically rely on synthetic augmentations of single images, we propose VET-DINO, a framework that leverages a unique characteristic of medical imaging: the availability of multiple standardized views from the same study. Using a series of clinical veterinary radiographs from the same patient study, we enable models to learn view-invariant anatomical structures and develop an implied 3D understanding from 2D projections. We demonstrate our approach on a dataset of 5 million veterinary radiographs from 668,000 canine studies. Through extensive experimentation, including view synthesis and downstream task performance, we show that learning from real multi-view pairs leads to superior anatomical understanding compared to purely synthetic augmentations. VET-DINO achieves state-of-the-art performance on various veterinary imaging tasks. Our work establishes a new paradigm for self-supervised learning in medical imaging that leverages domain-specific properties rather than merely adapting natural image techniques.

Foundation models (FMs) such as CLIP and SAM have recently shown great promise in image segmentation tasks, yet their adaptation to 3D medical imaging-particularly for pathology detection and segmentation-remains underexplored. A critical challenge arises from the domain gap between natural images and medical volumes: existing FMs, pre-trained on 2D data, struggle to capture 3D anatomical context, limiting their utility in clinical applications like tumor segmentation. To address this, we propose an adaptation framework called TAGS: Tumor Adaptive Guidance for SAM, which unlocks 2D FMs for 3D medical tasks through multi-prompt fusion. By preserving most of the pre-trained weights, our approach enhances SAM's spatial feature extraction using CLIP's semantic insights and anatomy-specific prompts. Extensive experiments on three open-source tumor segmentation datasets prove that our model surpasses the state-of-the-art medical image segmentation models (+46.88% over nnUNet), interactive segmentation frameworks, and other established medical FMs, including SAM-Med2D, SAM-Med3D, SegVol, Universal, 3D-Adapter, and SAM-B (at least +13% over them). This highlights the robustness and adaptability of our proposed framework across diverse medical segmentation tasks.

Friedreich ataxia (FRDA) is a rare, inherited progressive movement disorder for which there is currently no cure. The field urgently requires more sensitive, objective, and clinically relevant biomarkers to enhance the evaluation of treatment efficacy in clinical trials and to speed up the process of drug discovery. This study pioneers the development of clinically relevant, multidomain, fully objective composite biomarkers of disease severity and progression, using multimodal neuroimaging and background data (i.e., demographic, disease history, genetics). Data from 31 individuals with FRDA and 31 controls from a longitudinal multimodal natural history study IMAGE-FRDA, were included. Using an elasticnet predictive machine learning (ML) regression model, we derived a weighted combination of background, structural MRI, diffusion MRI, and quantitative susceptibility imaging (QSM) measures that predicted Friedreich ataxia rating scale (FARS) with high accuracy (R² = 0.79, root mean square error (RMSE) = 13.19). This composite also exhibited strong sensitivity to disease progression over two years (Cohen's d = 1.12), outperforming the sensitivity of the FARS score alone (d = 0.88). The approach was validated using the Scale for the assessment and rating of ataxia (SARA), demonstrating the potential and robustness of ML-derived composites to surpass individual biomarkers and act as complementary or surrogate markers of disease severity and progression. However, further validation, refinement, and the integration of additional data modalities will open up new opportunities for translating these biomarkers into clinical practice and clinical trials for FRDA, as well as other rare neurodegenerative diseases.

Visual grounding (VG) is the capability to identify the specific regions in an image associated with a particular text description. In medical imaging, VG enhances interpretability by highlighting relevant pathological features corresponding to textual descriptions, improving model transparency and trustworthiness for wider adoption of deep learning models in clinical practice. Current models struggle to associate textual descriptions with disease regions due to inefficient attention mechanisms and a lack of fine-grained token representations. In this paper, we empirically demonstrate two key observations. First, current VLMs assign high norms to background tokens, diverting the model's attention from regions of disease. Second, the global tokens used for cross-modal learning are not representative of local disease tokens. This hampers identifying correlations between the text and disease tokens. To address this, we introduce simple, yet effective Disease-Aware Prompting (DAP) process, which uses the explainability map of a VLM to identify the appropriate image features. This simple strategy amplifies disease-relevant regions while suppressing background interference. Without any additional pixel-level annotations, DAP improves visual grounding accuracy by 20.74% compared to state-of-the-art methods across three major chest X-ray datasets.

In many real-world applications, deployed models encounter inputs that differ from the data seen during training. Out-of-distribution detection identifies whether an input stems from an unseen distribution, while open-world recognition flags such inputs to ensure the system remains robust as ever-emerging, previously $unknown$ categories appear and must be addressed without retraining. Foundation and vision-language models are pre-trained on large and diverse datasets with the expectation of broad generalization across domains, including medical imaging. However, benchmarking these models on test sets with only a few common outlier types silently collapses the evaluation back to a closed-set problem, masking failures on rare or truly novel conditions encountered in clinical use. We therefore present $NOVA$, a challenging, real-life $evaluation-only$ benchmark of $\sim$900 brain MRI scans that span 281 rare pathologies and heterogeneous acquisition protocols. Each case includes rich clinical narratives and double-blinded expert bounding-box annotations. Together, these enable joint assessment of anomaly localisation, visual captioning, and diagnostic reasoning. Because NOVA is never used for training, it serves as an $extreme$ stress-test of out-of-distribution generalisation: models must bridge a distribution gap both in sample appearance and in semantic space. Baseline results with leading vision-language models (GPT-4o, Gemini 2.0 Flash, and Qwen2.5-VL-72B) reveal substantial performance drops across all tasks, establishing NOVA as a rigorous testbed for advancing models that can detect, localize, and reason about truly unknown anomalies.