Abstract Precise tumor margin delineation remains critical for effective cancer treatment, yet traditional histopathological methods often miss early molecular changes indicative of malignancy. We developed artificial intelligence-enhanced mid-infrared photothermal imaging (AI-enhanced MIP), a label-free platform for automated spatial profiling of patient tissues at submicron resolution by acquiring intrinsic molecular spectroscopic information at each pixel. Analysis of surgical specimens from patients with head and neck squamous cell carcinoma at various pathological stages revealed spectroscopically defined transition zones (SDTZs) near tumor margins—regions that appear histologically normal but exhibit distinct metabolic alterations. Single-nucleus RNA sequencing validation confirmed elevated SREBF1 and CPT1A expression along with epithelial-mesenchymal transition signatures in SDTZs. Machine learning-assisted identification of SDTZ molecular signatures further improved staging prediction accuracy by 3.7-fold compared with using tumor core assessment alone. For clinical translation, AI-enhanced MIP requires no tissue pretreatment and enables rapid assessment using four selective wavenumbers, providing more precise tumor margin definition during intraoperative evaluation that could reduce recurrence while avoiding treatment delays.
Ye et al. (2026) studied this question.