HERO：面向肿瘤学的鲁棒病理学编码器
原标题：HERO: Histology Encoder for Robust Representation in Oncology
AI 摘要
该论文提出 HERO（Histology Encoder for Robust Representation in Oncology），一个基于 ViT-G/14 的病理学基础模型，使用 DINO 和 iBOT 目标训练，并在约 57.5 万张临床全切片图像、5 亿个形态学均衡的切片上通过高分辨率 Gram anchoring 进行精炼。作者指出，现有病理基础模型在临床跨医院、扫描仪和染色方案使用时，编码器会混入采集因素，影响下游预测安全。HERO 在 PathoROB 和 PLISM 等基准上展现出最强的采集偏移鲁棒性，在 39 个切片级临床任务中平均排名第一，并在六个基准框架的等权分析中取得最佳平均排名。
正文节选
HERO: Histology Encoder for Robust Representation in Oncology Abstract Foundation models trained on large pathology image corpora now provide strong, transferable representations for computational pathology. Over the past few years a series of such models has been released, each trained on more slides than the last; on standard classification and segmentation benchmarks, the leading models are now separated by small margins. In clinical use, however, the foundation model is applied to images fro