The field of medical imaging and AI is moving towards a more nuanced understanding of clinical reasoning and diagnostic capabilities. Researchers are recognizing the limitations of existing benchmarks and datasets, which often prioritize classification accuracy over deeper reasoning abilities. As a result, there is a growing emphasis on developing more comprehensive and challenging evaluation frameworks that can assess the true clinical potential of AI models. This shift is driven by the need for more reliable and trustworthy AI systems that can support high-stakes diagnostic decisions. Notable papers in this area include:

• Beyond Classification Accuracy: Neural-MedBench, which introduces a new benchmark for probing the limits of multimodal clinical reasoning in neurology.
• EVLF-FM: Explainable Vision Language Foundation Model for Medicine, which presents a multimodal vision-language foundation model designed to unify broad diagnostic capability with fine-grain explainability.
• Radiology's Last Exam (RadLE), which evaluates the performance of frontier AI models against human experts and proposes a taxonomy of visual reasoning errors in radiology.
• Dolphin v1.0 Technical Report, which introduces a large-scale multimodal ultrasound foundation model that achieves state-of-the-art performance in various clinical tasks.
• MedQ-Bench, which establishes a perception-reasoning paradigm for language-based evaluation of medical image quality with Multi-modal Large Language Models (MLLMs).