构建新数据集并优化模型，提升癌细胞分裂图像检测准确率
Pan-Cancer mitotic figures detection and domain generalization: MIDOG 2025 Challenge
- 基于大规模数据训练，采用前沿方法提升检测能力
- 在常规与非典型分裂图像上分别取得0.8407和0.9107的高分
- 适合病理图像分析与跨域泛化研究者参考
本文介绍我们参与2025年有丝分裂域泛化（MIDOG）挑战赛的提交方案，该任务旨在通过组织病理学图像进行癌细胞有丝分裂检测，以辅助癌症预后判断。遵循‘苦教训’原则——强调数据规模胜过算法创新，我们公开发布两个新数据集，用于增强常规（Shen2024framework）与非典型有丝分裂（shen_2025_16780587）的训练数据。此外，我们采用了当前最新的训练方法，在测试集上实现了Track-1的F1-Score为0.8407，以及Track-2非典型有丝分裂细胞分类的平衡准确率为0.9107。
原文摘要 · Abstract (English)
This report details our submission to the Mitotic Domain Generalization (MIDOG) 2025 challenge, which addresses the critical task of mitotic figure detection in histopathology for cancer prognostication. Following the "Bitter Lesson"\cite{sutton2019bitterlesson} principle that emphasizes data scale over algorithmic novelty, we have publicly released two new datasets to bolster training data for both conventional \cite{Shen2024framework} and atypical mitoses \cite{shen_2025_16780587}. Besides, we implement up-to-date training methodologies for both track and reach a Track-1 F1-Score of 0.8407 on our test set, as well as a Track-2 balanced accuracy of 0.9107 for atypical mitotic cell classification.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。