Chromatin features as a label-free biomarker for intraoperative diagnosis of CNS tumors by AI-augmented third harmonic generation microscopy

Standard hematoxylin and eosin (H&E) histology remains the cornerstone for tumor diagnosis but is unavailable intraoperatively. The frozen section analysis also requires ∼30 min. Here, we present a compact label-free third harmonic generation (THG) microscope combined with a large-scale model Cell-SAM for rapid intraoperative imaging of CNS tumors. We provide direct mechanistic evidence that condensed DNA aggregates generate the nuclear THG contrast in human gliomas. We further validate chromatin condensation as a unique label-free biomarker across CNS tumors and lymphomas, revealing diagnostic features comparable to H&E histology. To handle the feature complexity arising from myelin and chromatin-derived contrast, we introduce Cell-SAM for segmenting key elements in THG images of CNS tumors and normal brain, with human-level accuracy (>98%). Statistical analysis of cellular and vascular densities further demonstrates the potential of THG for tumor typing. Together, compact THG imaging and Cell-SAM establish a mechanism-based and data-driven framework for real-time intraoperative pathology.