SRS158 - Real-time dynamic ICG quantification for objective intraoperative bowel perfusion assessment

Abstract Background Intraoperative assessment of bowel viability remains subjective, creating uncertainty in resection margins and risk of anastomotic leak. Although indocyanine green (ICG) fluorescence angiography is widely used, interpretation is largely qualitative and operator-dependent. Methods To facilitate objective intraoperative bowel perfusion assessment, we introduce an end-to-end workflow for real-time dynamic ICG quantification on videos. Firstly, the intestine is automatically segmented on the initial frame using HSV color-space thresholding to isolate bowel from background. A high-density grid of points is then initialized within the mask and robustly tracked using a state-of-the-art deep learning tracker (CoTracker) to handle physiologic motion and deformation. For each point, a fluorescence time-intensity curve is generated and smoothed, and quantitative metrics are computed, including maximum fluorescence (Fmax), time to peak (Tmax), time to half-peak (T1/2max), inflow gradient, and area under the curve (AUC). During the procedure, clinicians can overlay selected metrics as heatmaps on the live video, while per-point measurements and quantitative summaries are automatically saved for later review. Results The pipeline achieved stable tracking, clean time-intensity curves, and dense, spatially resolved perfusion maps. Metrics and heatmaps were generated automatically from raw videos, enabling objective comparison of segments and rapid identification of hypoperfused regions. Visualizations were intuitive and suitable for intraoperative review, supporting real-time interpretation. Conclusions This fully automated, deep learning-based system converts ICG signals into objective, high-granularity perfusion metrics and real-time dynamic heatmaps, reducing subjectivity and supporting intraoperative decision-making. Prospective evaluation is warranted to assess impact on outcomes and workflow integration.