Minimally Invasive, Label-Free, Point-of-Care Histopathological Diagnostic Platform of Malignant Tumors of the Female Reproductive System Based on Raman Spectroscopy and Machine Learning

Fast intraoperative histopathology is critical for optimal surgery in ovarian, endometrial and cervical cancers, yet frozen-section pathology is slow and resource-intensive. We obtained 4750 Raman spectra from 85 human gynecological tissue specimens spanning 19 histopathological classes. Spectra were preprocessed and classified with five machine-learning algorithms; performance was assessed by stratified train-test splits (70%:30%). Support-vector machines achieved 100% accuracy (AUC = 1.00) across all classes, outperforming random forest (96-99%) and k-nearest-neighbor (97-99%). Single-spectra acquisition required 30 s and automated prediction <8 s, enabling real-time decisions within 1 min. Raman-derived biochemical fingerprints highlighted subtype-specific alterations in nucleic acids, amino acids and collagen that are invisible to routine microscopy. Coupling Raman spectroscopy with machine learning yields an ultrarapid, label-free platform that accurately discriminates malignant, benign and premalignant lesions of the female reproductive tract at the point of care. The technology could reduce operative time, minimize repeat surgery and extend high-quality histopathology to low-resource settings.