Robust quantification of ICG fluorescence perfusion in neonatal bowel surgery via deep point tracking

Indocyanine green (ICG) fluorescence imaging is increasingly used for intraoperative bowel perfusion assessment in neonatal surgery. However, existing commercial solutions are limited to track few manually selected regions of interest (ROIs) and struggle with tissue motion and occlusion. We present a novel ICG quantification framework leveraging deep learning-based point tracking for robust perfusion analysis.

Our system employs CoTracker3, a state-of-the-art transformer-based point tracking model, to enable tracking of both user-specified ROIs and dense sampling across the entire visible tissue region. The framework computes comprehensive perfusion metrics including $F_{max}$ , $T_{max}$ , ingress gradient, $T_{1/2max}$ , and washout rate for each tracked region. For dense tissue analysis, dynamic perfusion heatmaps are generated through spatial interpolation with concave hull boundary reconstruction.

Preliminary validation on neonatal bowel perfusion videos demonstrates robust tracking performance under tissue motion and partial occlusion. The system successfully extracts perfusion metrics comparable to the commercial PerfusionTech system while enabling dense spatial perfusion mapping not available in existing solutions.

The proposed CoTracker3-based framework provides a feasible approach for quantitative ICG fluorescence analysis with improved tracking robustness and comprehensive spatial perfusion visualization, warranting further clinical validation studies.