Artificial intelligence has demonstrated significant potential in clinical decision-making; however, developing models capable of adapting to diverse real-world scenarios and performing complex diagnostic reasoning remains a major challenge. Existing medical multi-modal benchmarks are typically limited to single-image, single-turn tasks, lacking multi-modal medical image integration and failing to capture the longitudinal and multi-modal interactive nature inherent to clinical practice. To address this gap, we introduce MedAtlas, a novel benchmark framework designed to evaluate large language models on realistic medical reasoning tasks. MedAtlas is characterized by four key features: multi-round visual question answering (VQA), Joint reasoning of multiple modalities of medical images, multi-task integration, and high clinical fidelity. It supports four core tasks: open-ended multi-round VQA, closed-ended multi-round VQA, multi-image joint reasoning, and comprehensive disease diagnosis. Each case is derived from real diagnostic workflows and incorporates temporal interactions between textual medical histories and multiple imaging modalities, including CT, MRI, PET, ultrasound, X-ray, etc., requiring models to perform deep integrative reasoning across images and clinical texts. MedAtlas provides expert-annotated gold standards for all tasks. Furthermore, we propose two novel evaluation metrics: Stage Chain Accuracy (SCA) and Error Propagation Suppression Coefficient (EPSC). Benchmark results with existing multi-modal models reveal substantial performance gaps in multi-stage clinical reasoning. MedAtlas establishes a challenging evaluation platform to advance the development of robust and trustworthy medical AI.

Bladder cancer staging is crucial for operation planning and cancer assessment. Deep Convolutional Neural Networks (DCNNs) have been widely used to classify the bladder tumor images to identify cancer stages. However, the pure image-based deep learning methods over depend on the labeled data training and neglect the clinical priors. Human doctors judge the stage of a bladder tumor through checking whether the tumor infiltrating into bladder wall. The clinical priors of tumor infiltration are helpful to improve the DCNN-based bladder cancer staging and make the predictions coincide with the law of medicine. To involve clinical priors into deep learning for cancer staging, we propose a DCNN model with prior evidence to classify medical images of bladder tumors. Specifically, we measure the degree of tumor infiltrating into bladder wall to construct the prior evidence and integrate the prior evidence into the image-based prediction with evidential deep neural networks. We analyze the learning objective and prove that the prior evidences consistent with the ground truth will certainly reduce the prediction error and variance produced by image-based neural networks. The experiments on bladder cancer MR images datasets validate that involving prior evidences is effective to improve the DCNN-based cancer staging.