Computer vision improves biliary fluorescence contrast in retrospective cholecystectomy study

Background liver fluorescence can obscure extrahepatic biliary ducts during indocyanine green near-infrared fluorescence imaging in laparoscopic cholecystectomy. Researchers retrospectively analyzed 156 fluorescence-guided cases from a multi-institutional database of 928 procedures. Images were annotated for ducts, liver and instruments. A YOLOv11-based model selectively attenuated liver fluorescence, while a separate fusion approach used SAM3 segmentation to superimpose fluorescence-enhanced ducts on white-light liver images. The principal imaging measure was the duct-to-liver signal-to-background ratio (SBR).

Baseline SBR was 1.21 ± 0.69. Liver fluorescence attenuation increased SBR to 4.46 ± 3.06, and the fusion approach increased it to 3.07 ± 1.09; both comparisons had p < 0.0001. Multimodal detection models achieved precision of 0.911, recall of 0.866, an F1 score of 0.887 and mean average precision of 0.842. The authors reported better detection performance than previous models trained on white-light images alone.

These findings support technical feasibility for improving biliary image contrast. However, the retrospective analysis evaluated imaging and detection metrics rather than patient outcomes. The supplied report does not establish whether these approaches reduce bile duct injuries or improve operative decisions; prospective clinical evaluation would be needed before inferring such benefits.