arXiv 2603.270482026-06-23cs.CV

MOOZY: 一种以患者为先的计算病理学基础模型

MOOZY: A Patient-First Foundation Model for Computational Pathology

CSSE, Concordia University, Montr\' e al, Canada Mila -- Qu\' e bec AI Institute, Montr\' e al, Canada CHUM, Universit\' e de Montr\' e al, Montr\' e al, Canada IRIC, Universit\' e de Montr\' e al, Montr\' e al, Canada Universit\' e de Montr\' e al, Montr\' e al, Canada Canada CIFAR Chair, Polytechnique Montr\' e al, Montr\' e al, Canada Dept.\ of Pathology, McGill University, Montr\' e al, Canada

Yousef Kotp, Vincent Quoc-Huy Trinh, Christopher Pal, Mahdi S. Hosseini

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

提出以患者为基本单位的病理基础模型MOOZY，通过病例变换器建模同一患者多张切片依赖，结合多阶段自监督与低成本任务监督，在16个保留任务上优于PRISM。

摘要：

计算病理学需要能够跨多种临床任务迁移的全切片图像（WSI）基础模型，但当前方法仍以切片为中心，通常依赖私有数据和昂贵的配对报告监督，且未明确建模同一患者多个切片之间的关系。我们提出MOOZY，一种以患者为先的病理学基础模型，其中患者病例（而非单个切片）是表示的核心单元。MOOZY通过预训练期间的病例变换器显式建模同一患者所有切片之间的依赖关系，结合多阶段自监督与可扩展的低成本任务监督。在第一阶段，我们在77,134个公共切片特征网格上使用掩码自蒸馏预训练纯视觉切片编码器。在第二阶段，我们使用病例变换器和多任务监督将这些表示与临床语义对齐，监督来自56个公共数据集的333个任务，包括205个分类任务和128个生存任务（覆盖四个终点）。在16个保留任务上，相对于PRISM，MOOZY在宏加权F1、平衡准确率和宏加权ROC-AUC上分别提升+4.19%、+7.93%和+6.95%。MOOZY参数高效，仅85.77M参数，比GigaPath小14倍。这些结果表明，患者级预训练产生可迁移的嵌入，为可扩展的以患者为先的组织病理学基础模型提供了途径。

英文摘要：

Computational pathology needs whole-slide image (WSI) foundation models that transfer across diverse clinical tasks, yet current approaches remain largely slide-centric, often depend on private data and expensive paired-report supervision, and do not explicitly model relationships among multiple slides from the same patient. We present MOOZY, a patient-first pathology foundation model in which the patient case, not the individual slide, is the core unit of representation. MOOZY explicitly models dependencies across all slides from the same patient via a case transformer during pretraining, combining multi-stage self-supervision with scaled low-cost task supervision. In Stage 1, we pretrain a vision-only slide encoder on 77,134 public slide feature grids using masked self-distillation. In Stage 2, we align these representations with clinical semantics using a case transformer and multi-task supervision over 333 tasks from 56 public datasets, including 205 classification and 128 survival tasks across four endpoints. Across sixteen held-out tasks, MOOZY improves macro weighted F1, balanced accuracy, and macro weighted ROC-AUC relative to PRISM by +4.19\%, +7.93\%, and +6.95\%, respectively. MOOZY is also parameter efficient with 85.77M parameters, 14$\times$ smaller than GigaPath. These results suggest that patient-level pretraining yields transferable embeddings, providing a path toward scalable patient-first histopathology foundation models.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。