Scholarlink

<p>To address the clinical challenge of postoperative lower extremity deep vein thrombosis (LEDVT) in endometrial cancer care, this study establishes an explainable machine learning framework for personalized risk prediction. Utilizing perioperative data from 841 patients across multiple centers, we evaluated 26 machine learning algorithms combined with diverse data augmentation techniques. The Support Vector Machine (SVM) model emerged as the most robust architecture, refined through recursive feature elimination to a concise four-variable set comprising postoperative D-dimer, age, fibrinogen, and clinical stage. The model demonstrated superior discriminative performance, achieving an area under the curve (AUC) of 0.828 in internal validation and 0.819 in an independent external cohort. To bridge the gap between “black-box" AI and clinical trust, we integrated SHapley Additive exPlanations (SHAP) to quantify individual feature contributions, revealing non-linear associations such as the critical risk threshold for D-dimer levels. Finally, a web-based decision support interface was implemented to provide real-time, interpretable risk assessments. By combining high predictive accuracy with transparent decision logic, this approach offers a precise tool for identifying high-risk patients and optimizing postoperative management in endometrial cancer care.</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

Personalised thrombo-embolic risk prediction after endometrial cancer surgery: an explainable AI approach using SHAP

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月27日

查看全文
  • Qing Zhou,
  • Fudan Liu,
  • Jiaxiu Li,
  • Qiyin Zhou,
  • Xiaorong Xiang,
  • Cheng Dai,
  • Lianli Hen,
  • ShouLi Dao,
  • Mingjie Zhang,
  • Bowen Gao,
  • Yuxi Li,
  • Lin Xu,
  • Yonghu Chang,
  • Donghong Wang

摘要

To address the clinical challenge of postoperative lower extremity deep vein thrombosis (LEDVT) in endometrial cancer care, this study establishes an explainable machine learning framework for personalized risk prediction. Utilizing perioperative data from 841 patients across multiple centers, we evaluated 26 machine learning algorithms combined with diverse data augmentation techniques. The Support Vector Machine (SVM) model emerged as the most robust architecture, refined through recursive feature elimination to a concise four-variable set comprising postoperative D-dimer, age, fibrinogen, and clinical stage. The model demonstrated superior discriminative performance, achieving an area under the curve (AUC) of 0.828 in internal validation and 0.819 in an independent external cohort. To bridge the gap between “black-box" AI and clinical trust, we integrated SHapley Additive exPlanations (SHAP) to quantify individual feature contributions, revealing non-linear associations such as the critical risk threshold for D-dimer levels. Finally, a web-based decision support interface was implemented to provide real-time, interpretable risk assessments. By combining high predictive accuracy with transparent decision logic, this approach offers a precise tool for identifying high-risk patients and optimizing postoperative management in endometrial cancer care.