New AI framework matches scans to standard pathology tests

Researchers have developed AFN-DeSeg, a deep learning framework that correlates directly with hematoxylin and eosin pathology in identifying key diagnostic areas of high-grade serous ovarian carcinoma. The work addresses a limitation of two-photon autofluorescence microscopy, a promising method for rapid, label-free tissue assessment, by improving the visibility of nuclei, a cancer biomarker historically difficult to detect due to naturally low contrast. By unifying image denoising and nuclear segmentation, AFN-DeSeg recovers diagnostic-grade nuclear features and “bridges the gap between label-free TPAF imaging and H&E-based histology,” establishing a computational foundation for potential intraoperative optical biopsy. Standard segmentation approaches for identifying cancerous cells often depend on clear boundaries, yet two-photon autofluorescence (TPAF) imaging presents challenges due to the negative contrast of nuclei and a low signal-to-noise ratio that obscures their morphology. To overcome these limitations, researchers constructed a composite dataset using both public fluorescence microscopy images and in-house ovarian cancer TPAF images, addressing the scarcity of paired TPAF data needed for training…