Abstract
Background: Deep learning algorithms have shown promise for real-time tissue classification during laparoscopic procedures, potentially improving surgical safety and reducing complications. This systematic review evaluates the current evidence.
Methods: We searched PubMed, EMBASE, and IEEE Xplore for studies evaluating deep learning for intraoperative tissue classification in laparoscopic surgery through December 2025. Studies reporting classification accuracy on surgical video were included. Meta-analysis was performed using random-effects models. Risk of bias was assessed with QUADAS-2.
Results: Twenty-four studies comprising 18,400 surgical images and 620 procedures were included. Pooled accuracy for tissue classification was 91.3% (95% CI: 88.7–93.6%). Performance was highest for large bowel identification (94.1%) and lowest for neural structure recognition (85.2%). Twelve studies reported real-time performance (>24 fps). Significant heterogeneity was present (I²=78%).
Conclusions: Deep learning achieves high accuracy for intraoperative tissue classification in laparoscopic surgery. Translation to clinical practice requires prospective validation and standardized reporting frameworks.
Full manuscript text available in the uploaded document file. Authors may provide HTML-formatted full text for inline display.