📄 中文摘要
卵巢癌的组织病理亚型研究对个性化有效治疗策略的制定具有重要价值。然而，英国病理部门面临日益增加的诊断工作负担，这促使了人工智能方法的兴起。传统方法依赖于预计算的冻结图像特征，而近年来的进展则转向端到端特征提取，尽管提高了准确性，但在训练时显著降低了可扩展性，并增加了实验的时间成本。研究提出了一种新的方法，通过对比学习和原型学习，结合预计算的图像块特征，实现卵巢癌组织病理图像的亚型分类与定位。
MB-DSMIL-CL-PL：基于对比学习和原型学习的可扩展弱监督卵巢癌亚型分类与定位方法，使用冻结的图像块特征
📄 English Summary
MB-DSMIL-CL-PL: Scalable Weakly Supervised Ovarian Cancer Subtype Classification and Localisation Using Contrastive and Prototype Learning with Frozen Patch Features
The study of histopathological subtypes is crucial for personalizing effective treatment strategies for ovarian cancer. Increasing diagnostic workloads in UK pathology departments have led to a rise in AI approaches. Traditional methods have relied on pre-computed, frozen image features, while recent advances have shifted towards end-to-end feature extraction, improving accuracy but significantly reducing scalability during training and increasing experimental time. A new approach is proposed for subtype classification and localization in ovarian cancer histopathology images using contrastive and prototype learning with pre-computed patch features.