Artificial Intelligence for Free Flap Monitoring in Reconstructive Surgery: A Review

Abstract

Free flap reconstruction is widely used in reconstructive surgery, but postoperative vascular compromise remains a major cause of flap failure. Early detection of arterial insufficiency and venous congestion is essential for timely intervention and flap salvage. Conventional monitoring is limited by observer variability, flap accessibility, and assessment frequency. Artificial intelligence (AI) may support more consistent and continuous surveillance. This review examines the diagnostic performance, clinical utility, implementation challenges, and research priorities of AI-based postoperative free flap monitoring. Applications include smartphone photography, serial clinical images, thermal imaging, near-infrared spectroscopy, hyperspectral imaging, and other monitoring signals. Tasks encompass classification, segmentation, severity grading, and longitudinal prediction, distinguishing detection of existing perfusion abnormalities from prediction of subsequent compromise. Emerging studies report promising performance in identifying impaired perfusion; however, diagnostic accuracy alone does not establish clinical effectiveness. Limitations include small or single-center cohorts, image-level data leakage, class imbalance, inconsistent reference standards, limited external validation, and insufficient evidence of improved flap salvage. Performance may vary with flap location, tissue characteristics, skin pigmentation, image quality, and monitoring protocols. Intraoral and buried flaps pose additional accessibility challenges. Safe implementation requires standardized data acquisition, transparent alert thresholds, clinician oversight, privacy protection, and clear escalation pathways. AI may complement established clinical assessment, but routine adoption requires multicentre validation and prospective evidence of clinical benefit. Future studies should prioritize patient-level performance, detection time, false-negative events, unnecessary alerts, workload, flap salvage, and patient outcomes.