New Halsted vision-language model trained on 650,000+ surgical videos aims to make AI-powered surgery mapping accessible to practicing surgeons

Halsted, a vision-language model trained on the Halsted Surgical Atlas (HSA), achieves state-of-the-art performance in mapping surgical behavior from video across eight surgical specialties

HSA comprises over 650,000 annotated videos built through an iterative self-labeling framework, making it one of the most comprehensive surgical video libraries available

Researchers publicly released HSA-27k, a 27,000-video subset for benchmarking, enabling wider access and comparison for the research community

The model offers improved comprehensiveness and computational efficiency compared to previous approaches while remaining practical for actual clinical use by surgeons