Impact: Multimodal LLMs, combined with segmentation-derived metrics and clinical data, enable the generation of structured, quantitative reports, potentially enhancing diagnostic support, triage efficiency and patient communication, particularly valuable in under-resourced settings, where MRI staff shortages cause delays in diagnosis and treatment.

Impact: MMRQA integrates signal metrics with multimodal LLMs to deliver interpretable MRI quality assessments, enabling rapid artifact detection and clinical decision-making in data-scarce environments, potentially reducing diagnostic errors and optimizing protocols across diverse MR acquisitions.

Impact: The proposed automated foundation model overcomes the major barriers of manual prompting and annotation scarcity. It enables LV scar volume to become a practical, standard prognostic metric, accelerating personalized risk stratification in cardiovascular medicine.

Impact: We show how large language models can provide a novel route for injecting prior neuroanatomical knowledge into connectomics studies, with demonstrated benefits for improving the sensitivity of tractography filtering in a mechanistic model of Alzheimer’s disease.

Impact: Vision-language AI can enhance prostate MRI interpretation by integrating automated lesion detection, quantitative analysis, and structured reporting. This approach can reduce inter-reader variability while enabling standardized, reproducible, and efficient prostate cancer diagnosis, communication, and data-driven research integration.

Impact: Knowledge-enhanced LLMs show diagnostic performance comparable to experienced radiologists in glioma classification and improve junior radiologists’ accuracy. These findings suggest LLMs may serve as valuable decision-support tools, though limitations in certain grading scenarios underscore the necessity of radiologist oversight.

Impact: This work introduces vision-language foundation models into fast MRI reconstruction, demonstrating that enforcing semantic consistency improves perceptual quality and structural fidelity. The approach integrates linguistic understanding into image reconstruction, enriching the representational space and promoting multimodal reconstruction in medical imaging.