Scholarlink

Objective <p>To investigate the availability of machine learning based on MRI radiomics for predicting key genetic mutations of consensus molecular classification in muscle-invasive bladder cancer (MIBC).</p> Methods <p>Preoperative MRI, postoperative DNA sequencing data and prognosis information of MIBC patients (stage T2–T4aN0M0) were retrospectively analyzed. Radiomic features were extracted from lesions on T2-weighted, diffusion-weighted images and apparent diffusion coefficient maps. The Least absolute shrinkage and selection operate with tenfold cross-validation was used to generate radiomics signatures. Random forest (RF), support vector machine (SVM), and logistic regression (LR) models were trained and internally validated by leave-one-out. The areas under the receiver operator characteristic curve (AUC), balanced accuracy (BACC) and Matthews correlation coefficient (MCC) were calculated to assess classifier performance. Disease-free survival (DFS) was analyzed by Kaplan–Meier test.</p> Results <p>Among 91participants, TP53, ERBB2, RB1, FGFR3 mutation and TP53 co-mutation with RB1 were 60, 23, 19, 15 and 16. Compared to FGFR3 wild type, mutated tumors had a higher proportion of histologically low-grade lesions (26.7% vs. 1.4%, <i>p</i> = 0.003). The AUCs of RF model for predicting ERBB2 and RB1 mutation were 0.89 and 0.90, BACCs were 0.77 and 0.84, MCCs were 0.51 and 0.57. While for predicting TP53, FGFR3 mutation and co-mutation, SVM performed better, AUCs reached 0.86,0.94 and 0.91, BACCs were 0.75, 0.60 and 0.79, MCCs were 0.51, 0.60 and 0.79. The co-mutation from sequencing (<i>p</i> = 0.009) and SVM predicting (<i>p</i> &lt; 0.001) significantly associated with poor DFS.</p> Conclusion <p>Radiomic signatures can predict key genetic mutations well in MIBC, and support references to clinical decision.</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

Radiogenomics of muscle-invasive bladder cancer: machine learning for predicting prognostically relevant genetic mutations

  • 期刊论文

  • 发表日期：

    2026年3月12日

查看全文
  • Hua Wei,
  • Zhichang Fan,
  • Yan Li,
  • Wenqiao Zheng,
  • Xiaoyue Zhang,
  • Zeke Chen,
  • Wenxin Li,
  • Jiangfeng Du,
  • Xiaochun Wang

摘要

Objective

To investigate the availability of machine learning based on MRI radiomics for predicting key genetic mutations of consensus molecular classification in muscle-invasive bladder cancer (MIBC).

Methods

Preoperative MRI, postoperative DNA sequencing data and prognosis information of MIBC patients (stage T2–T4aN0M0) were retrospectively analyzed. Radiomic features were extracted from lesions on T2-weighted, diffusion-weighted images and apparent diffusion coefficient maps. The Least absolute shrinkage and selection operate with tenfold cross-validation was used to generate radiomics signatures. Random forest (RF), support vector machine (SVM), and logistic regression (LR) models were trained and internally validated by leave-one-out. The areas under the receiver operator characteristic curve (AUC), balanced accuracy (BACC) and Matthews correlation coefficient (MCC) were calculated to assess classifier performance. Disease-free survival (DFS) was analyzed by Kaplan–Meier test.

Results

Among 91participants, TP53, ERBB2, RB1, FGFR3 mutation and TP53 co-mutation with RB1 were 60, 23, 19, 15 and 16. Compared to FGFR3 wild type, mutated tumors had a higher proportion of histologically low-grade lesions (26.7% vs. 1.4%, p = 0.003). The AUCs of RF model for predicting ERBB2 and RB1 mutation were 0.89 and 0.90, BACCs were 0.77 and 0.84, MCCs were 0.51 and 0.57. While for predicting TP53, FGFR3 mutation and co-mutation, SVM performed better, AUCs reached 0.86,0.94 and 0.91, BACCs were 0.75, 0.60 and 0.79, MCCs were 0.51, 0.60 and 0.79. The co-mutation from sequencing (p = 0.009) and SVM predicting (p < 0.001) significantly associated with poor DFS.

Conclusion

Radiomic signatures can predict key genetic mutations well in MIBC, and support references to clinical decision.