Scholarlink

Prostate cancer is one of the most prevalent and dangerous diseases in men, affecting 1.4 million individuals annually. Although whole-slide images (WSIs) reveal histopathological features at the cellular level, computational tools for predicting biochemical recurrence in prostate cancer patients remain limited. Recent developments in foundation pathology models, which are pretrained on large-scale, unlabeled datasets in a self-supervised manner, offer promising opportunities for WSI analysis. However, these models have so far been evaluated primarily on cancer subtyping, grading, or survival analysis, and their performance in predicting biochemical recurrence has not yet been investigated. In this study, we present a comprehensive evaluation of five pathology foundation models (CTransPath, UNI, ProvGigaPath, CONCH, and TITAN) as feature extractors for patches tiled from WSIs in the biochemical recurrence task, thereby bridging the knowledge gap regarding these models’ applicability and assessing their effectiveness for this critical prediction.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predicting Biochemical Recurrence of Prostate Cancer from Whole-Slide Images: A Benchmark of Pathology Foundation Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Doanh C. Bui,
  • Hoai Luan Pham,
  • Vu Trung Duong Le,
  • Yasuhiko Nakashima

摘要

Prostate cancer is one of the most prevalent and dangerous diseases in men, affecting 1.4 million individuals annually. Although whole-slide images (WSIs) reveal histopathological features at the cellular level, computational tools for predicting biochemical recurrence in prostate cancer patients remain limited. Recent developments in foundation pathology models, which are pretrained on large-scale, unlabeled datasets in a self-supervised manner, offer promising opportunities for WSI analysis. However, these models have so far been evaluated primarily on cancer subtyping, grading, or survival analysis, and their performance in predicting biochemical recurrence has not yet been investigated. In this study, we present a comprehensive evaluation of five pathology foundation models (CTransPath, UNI, ProvGigaPath, CONCH, and TITAN) as feature extractors for patches tiled from WSIs in the biochemical recurrence task, thereby bridging the knowledge gap regarding these models’ applicability and assessing their effectiveness for this critical prediction.