The field of medical imaging and vision-language models is rapidly evolving, with a focus on developing more accurate and efficient models for tasks such as image segmentation, disease diagnosis, and report generation. Recent work has emphasized the importance of domain-specific foundation models, which can provide a unified framework for multiple clinical tasks and improve performance on downstream tasks. Additionally, there is a growing interest in vision-language pretraining, which enables representation learning from large-scale image-text pairs without relying on expensive manual annotations. Notable papers in this area include Mammo-FM, which introduces a breast-specific foundational model for integrated mammographic diagnosis, prognosis, and reporting, and STAMP, which resolves the trilemma in MLLM-based segmentation with simultaneous textual mask prediction. Other noteworthy papers include MedGrounder, which achieves strong zero-shot transfer and outperforms REC-style and grounded report generation baselines, and Panel2Patch, which enables granularity-aware pretraining and achieves substantially better performance with less pretraining data.