Scholarlink

Purpose <p>This study aims to develop and validate a multi-regional radiomics machine learning model integrating the spatial heterogeneity features of both the primary gross tumor volume (GTVp) and metastatic lymph nodes (GTVn) to explore its potential as a non-invasive tool for estimating the 5-year recurrence risk of HPV-positive oropharyngeal carcinoma (OPC) patients.</p> Methods <p>This study included 716 HPV + OPC patients from three independent datasets (training set: 390; internal validation set: 166; independent test set: 160). PyRadiomics was used to extract radiomic features from each target volume. Ten classical machine learning models were constructed, and the SHAP algorithm was applied to quantify feature contributions and optimize core variables. A clinical-radiomics model based solely on GTVp and a clinical-radiomics model based on a feature-level fusion strategy (GTVp&amp;n) were established respectively.</p> Results <p>The XGBoost model based on the GTVp&amp;n strategy showed the best performance, with AUCs of 0.907, 0.852, and 0.835 in the training, internal validation, and independent test set, respectively—significantly higher than those of the single GTVp model and the GTVpn (fused target volume) model. A nomogram incorporating clinical factors demonstrated good discrimination, calibration, and clinical utility across all datasets. NRI (0.242) and IDI (0.097) analyses confirmed that this combined model significantly improved risk reclassification capability compared to the GTVp-based clinical-radiomics model.</p> Conclusion <p>A multi-regional radiomics model integrating features from both the primary lesion and cervical lymph nodes significantly enhances the predictive performance for recurrence risk in HPV + OPC. This model facilitates individualized recurrence risk stratification and may generate research hypotheses for future exploration of personalized de-escalation or intensified treatment strategies, as well as supporting individualized follow-up management.</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

Multi-regional radiomics based on planning CT and information complementarity between GTVp and GTVn: an explainable model for predicting 5-year recurrence risk in HPV-positive oropharyngeal cancer

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月8日

查看全文
  • Chunsheng Wang,
  • Mingjun Ding,
  • Xiang Cao,
  • Yizhi Ge,
  • Shanliang Hu,
  • Jianguo Zhang,
  • Linzhi Han,
  • Chenjing Zhu,
  • Xia He

摘要

Purpose

This study aims to develop and validate a multi-regional radiomics machine learning model integrating the spatial heterogeneity features of both the primary gross tumor volume (GTVp) and metastatic lymph nodes (GTVn) to explore its potential as a non-invasive tool for estimating the 5-year recurrence risk of HPV-positive oropharyngeal carcinoma (OPC) patients.

Methods

This study included 716 HPV + OPC patients from three independent datasets (training set: 390; internal validation set: 166; independent test set: 160). PyRadiomics was used to extract radiomic features from each target volume. Ten classical machine learning models were constructed, and the SHAP algorithm was applied to quantify feature contributions and optimize core variables. A clinical-radiomics model based solely on GTVp and a clinical-radiomics model based on a feature-level fusion strategy (GTVp&n) were established respectively.

Results

The XGBoost model based on the GTVp&n strategy showed the best performance, with AUCs of 0.907, 0.852, and 0.835 in the training, internal validation, and independent test set, respectively—significantly higher than those of the single GTVp model and the GTVpn (fused target volume) model. A nomogram incorporating clinical factors demonstrated good discrimination, calibration, and clinical utility across all datasets. NRI (0.242) and IDI (0.097) analyses confirmed that this combined model significantly improved risk reclassification capability compared to the GTVp-based clinical-radiomics model.

Conclusion

A multi-regional radiomics model integrating features from both the primary lesion and cervical lymph nodes significantly enhances the predictive performance for recurrence risk in HPV + OPC. This model facilitates individualized recurrence risk stratification and may generate research hypotheses for future exploration of personalized de-escalation or intensified treatment strategies, as well as supporting individualized follow-up management.