Scholarlink

<p>Survival assessment for oral squamous cell carcinoma (OSCC) remains a significant clinical challenge. This study develops novel artificial intelligence (AI) platforms for assessing overall survival in OSCC patients based on 240 whole-slide images from multicenter cohorts. A comprehensive evaluation is conducted on four convolutional neural network architectures under two distinct deep learning (DL) training paradigms: supervised DL with precise annotations (PathS model, c-index = 0.809), and weakly supervised DL using slide-level labels without manual annotations (c-index = 0.707). Gradient-weighted class activation mapping reveals novel AI-based prognostic insights to simultaneously identify tumor cells and tumor-infiltrating immune cells as key predictive features. Additionally, our platform achieved significantly improved accuracy compared to conventional clinical signatures (CS model, c-index = 0.721). Furthermore, the clinical potential is enhanced through the development of a multimodal nomogram combining PathS signatures with CS (c-index = 0.817), representing a substantial advancement in personalized survival assessment for OSCC patients.</p>

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

Tumor cell- and infiltrating immune cell-based supervised learning artificial intelligence multimodal platform for tumor prognosis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月14日

查看全文
  • Xin-Jia Cai,
  • Chao-Ran Peng,
  • Chuan-Yang Ding,
  • Ying-Ying Cui,
  • Li Gao,
  • Zhi-Xiu Xu,
  • Long Li,
  • Jian-Yun Zhang,
  • Tie-Jun Li

摘要

Survival assessment for oral squamous cell carcinoma (OSCC) remains a significant clinical challenge. This study develops novel artificial intelligence (AI) platforms for assessing overall survival in OSCC patients based on 240 whole-slide images from multicenter cohorts. A comprehensive evaluation is conducted on four convolutional neural network architectures under two distinct deep learning (DL) training paradigms: supervised DL with precise annotations (PathS model, c-index = 0.809), and weakly supervised DL using slide-level labels without manual annotations (c-index = 0.707). Gradient-weighted class activation mapping reveals novel AI-based prognostic insights to simultaneously identify tumor cells and tumor-infiltrating immune cells as key predictive features. Additionally, our platform achieved significantly improved accuracy compared to conventional clinical signatures (CS model, c-index = 0.721). Furthermore, the clinical potential is enhanced through the development of a multimodal nomogram combining PathS signatures with CS (c-index = 0.817), representing a substantial advancement in personalized survival assessment for OSCC patients.