Topic: Surgical Data Science in the Era of Large Multimodal Models

Surgical data science plays a pivotal role in advancing modern interventional medicine by leveraging surgical videos and clinical data to improve surgical training, intraoperative decision-making, and patient safety. Conventionally, artificial intelligence in surgery relied on single-modality data and task-specific algorithms. However, the emergence of foundation models (FMs), including large language models (LLMs), vision-language models (VLMs), and vision-language-action models (VLAs), is profoundly reshaping this field.

We are now entering a transformative era where large multimodal models (LMMs) can simultaneously integrate and reason over complex surgical videos, medical images, and textual data. This paradigm shift enables deeper comprehension of surgical scenes and advanced reasoning capabilities. For instance, VLMs offer opportunities to align visual surgical events with natural language descriptions. The integration of visual and textual surgical data empowers intelligent systems to perform highly complex tasks, bridging the gap between visual perception and semantic understanding in surgical data science. Realizing this potential in clinical practice will also require rigorous validation of model performance, reliability, interpretability, and seamless integration into surgical workflows.

To highlight these cutting-edge advancements, this special issue invites original research articles that explore the intersection of surgical data science and large multimodal models. We particularly welcome submissions including, but not limited to, the following topics:

● Vision-language models and large multimodal models for surgical data understanding;

● Vision-language pre-training for surgical data analysis;

● Surgical action recognition and anticipation;

● Surgical phase and instrument recognition, activity boundary detection;

● Surgical skill assessment using multimodal surgical data;

● Video-text cross-modal retrieval and surgical video captioning;

● Interpretability and explainability of vision-language models;

● Learning from limited, sparse, imbalanced, weakly paired, or noisy surgical vision-language data;

● Development, validation, and comparative evaluation of AI-generated surgical reports, captions, and textual annotations against expert reference standards;

● Human–robot collaboration and multimodal decision support using surgical data;

● Trustworthy large multimodal models: robustness, uncertainty estimation, bias, privacy, and clinical validation.