1254-Pan-Cancer Tumour Segmentation in Whole-Slide Images

This research introduces a universal deep learning model designed for the automatic segmentation of tumors in digital histopathological images across various cancer types. By training on over 20,000 whole-slide images, the researchers developed a system capable of identifying cancerous regions in colorectal, lung, prostate, and endometrial carcinomas, as well as types not included in the training data like breast cancer. The study demonstrates that a single pan-cancer model performs as effectively as specialized versions while maintaining robustness across different slide scanners and laboratory preparations. Performance was evaluated using the Dice similarity coefficient, showing high accuracy in most solid tumor resections, though challenges remained with fragmented early-stage bladder samples. Ultimately, the findings suggest that automated AI tools can significantly streamline diagnostic pathology by providing consistent, large-scale tumor delineation.References:Skrede O J, Pradhan M, Isaksen M X, et al. Generalisation of automatic tumour segmentation in histopathological whole-slide images across multiple cancer types[J]. npj Precision Oncology, 2026.前往小宇宙评论区与主播互动