External validation and clinical readiness of intraoperative video AI in general surgery: a systematic review

Intraoperative video artificial intelligence (AI) has advanced rapidly in general surgery, however, its dependable clinical use remains uncommon. This systematic review evaluated the use of intraoperative video AI in general surgery with an emphasis on external validation, clinical readiness, and translational maturity. Studies evaluating AI systems using intraoperative video or video-derived visual input in general surgery were systematically reviewed. Eligible applications included workflow recognition, anatomy-oriented scene understanding, instrument recognition, event detection, difficulty assessment, predictive analytics, and skill evaluation. Evidence was synthesized narratively using a structured appraisal focused on validation rigor, reporting quality, workflow relevance, and implementation-oriented characteristics of the studies. Twenty-two studies were included in the analysis. The literature is dominated by retrospective, internally validated studies. Phase, step, and workflow recognition were the most extensively investigated domains. True external validation is uncommon, prospective or deployment-grade evaluation is rare, and technical maturity generally outpaces translational maturity across task domains. Intraoperative video AI in general surgery now spans a broad range of credible technical applications. However, limited external validation, scarce prospective evaluation, and insufficient evidence for workflow-integrated performance continue to constrain its dependable clinical use. Future progress will depend on stronger evidence for generalizability, usability, and clinically meaningful benefit in real-world surgical practice.