AI-Based Blood Perfusion Assessment in Digestive Surgery Using Indocyanine Green Time–Fluorescence Curves

An accurate intraoperative assessment of tissue perfusion in digestive surgery is critical to prevent complications such as anastomotic leakage, an event in which a surgical connection between organs fails to heal properly. However, current fluorescence angiography with indocyanine green (ICG) is still mostly qualitative and affected by surgeon expertise. This study proposes an artificial intelligence (AI)-based framework for quantitative perfusion evaluation using ICG fluorescence imaging. The system combines computer vision (CV) and machine learning (ML) to extract dynamic fluorescence profiles from intraoperative videos, cluster perfusion patterns, and classify tissue, focusing on the quality-related aspects of perfusion, as ideal and non-ideal perfusion. The framework was validated on 20 near-infrared ICG videos acquired during robotic colorectal procedures (including 3 cases with postoperative anastomotic leakage). After the preprocessing stage including stabilization and segmentation, regions of interest (ROIs) were tracked by extracting ICG time–fluorescence curves. The curves were clustered with k-medoids approach to separate optimal from non-ideal perfusion patterns. The resulting labels were used to train a logistic regression classifier, evaluated with stratified 5-fold cross-validation. The unsupervised step achieved a normalized Silhouette score of 0.77, indicating well-separated perfusion clusters. The supervised classifier reached a mean accuracy of 97±2 % in distinguishing optimal from suboptimal perfusion patterns. These findings demonstrate the potential of AI-enhanced fluorescence imaging to provide quantitative intraoperative decision support, reducing reliance on subjective interpretation and improving surgical precision. This approach advances fluorescence-guided surgery, offering a scalable, data-driven solution to minimize surgical complications.