Head and neck cancer surgery and label-free fluorescence lifetime imaging (FLIm)

We pair multispectral label-free FLIm with a data-centric AI model to generate real-time tumor-margin maps during head-and-neck cancer surgery. In a 92-patient study, confident-learning pruning and gradient-boosted trees achieved an AUC of 0.94 under leave-one-patient-out validation, with real time overlays rendered in 41 milliseconds latency. Borderline predictions tracked metabolic transitions in spectral channels associated with NADH- and FAD autofluorescence and remained robust across anatomical sites and HPV status. The framework offers accurate, clinical workflow-compatible guidance for surgeons and promising safer resections.