文献速递泌尿生殖腹部MR

基于多参数MRI的可解释机器学习模型预测去势抵抗性前列腺癌风险

mpMRI-based interpretable machine learning model for predicting castration-resistant prostate cancer risk.
PubMed 文献 · Eur J Radiol · 2025-11-20

本研究回顾性纳入两家医院2014年1月至2024年10月ADT前接受mpMRI的180例晚期前列腺癌患者,按至CRPC时间分为极高危(<1年,n=49)、高危(1-4年,n=65)和低危(>4年,n=66)组,构建8种机器学习分类器及堆叠集成模型,mpMRI-临床联合模型优于单纯mpMRI模型(AUC 0.84;95%CI 0.73, 0.92),堆叠模型内部测试集AUC 0.89(95%CI 0.81, 0.93)、外部测试集AUC 0.82(95%CI 0.72, 0.89),低危组AUC 0.89(95%CI 0.77, 0.97)。

要点

  • 按至CRPC时间分层:极高危<1年49例、高危1-4年65例、低危>4年66例。
  • mpMRI-临床联合模型AUC 0.84(95%CI 0.73, 0.92),优于单纯mpMRI模型。
  • 堆叠模型内部测试集AUC 0.89(95%CI 0.81, 0.93),外部测试集AUC 0.82(95%CI 0.72, 0.89)。

核心结论

  • 堆叠模型对晚期前列腺癌CRPC进展风险表现出良好的预测能力,有助于临床可操作的风险分层干预。

研究设计与样本

  • 回顾性研究,纳入两家医院2014年1月至2024年10月ADT前接受mpMRI的180例晚期前列腺癌患者,年龄均值72岁(范围51-90岁)。

诊断标准与关键数值

  • 按至CRPC时间分层:极高危<1年(n=49)、高危1-4年(n=65)、低危>4年(n=66);mpMRI-临床联合模型AUC 0.84(95%CI 0.73, 0.92);堆叠模型内部测试集AUC 0.89(95%CI 0.81, 0.93),外部测试集AUC 0.82(95%CI 0.72, 0.89);低危组堆叠模型AUC 0.89(95%CI 0.77, 0.97)。

临床应用与阅片要点

  • 原文未据此给出推荐。

局限性

  • 原文未报告局限性。

以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。

英文原文(摘要)· 点击展开

RATIONALE AND OBJECTIVES: Androgen deprivation therapy (ADT) efficacy exhibits significant interindividual heterogeneity in advanced prostate cancer (PCa). This study aim to develop and validate a machine learning model based on multiparametric MRI (mpMRI) to predict the risk of Castration-Resistant Prostate Cancer (CRPC) progression in advanced PCa.

MATERIALS AND METHODS: In this retrospective study, 180 advanced PCa patients who underwent mpMRI before ADT were collected from two hospitals between January 2014 and October 2024. Radiomic features were selected by using variance threshold, Least Absolute Shrinkage and Selection Operator and recursive feature elimination. Random forest and logistic regression model selected clinical features. We developed eight machine learning classifiers, a stacking ensemble model by integrating the optimally performing classifier. Model performance was evaluated using confusion matrices, accuracy, precision, recall, F1-score, and area under curve (AUC). SHapley Additive exPlanations was employed for both global and local model interpretability. Intergroup differences were analyzed using one-way Analysis of Variance or Kruskal-Wallis tests.

RESULTS: 180 patients (mean age 72, range 51-90 years) were stratified by the time to CRPC into groups: very high-risk (<1 year, n = 49), high-risk (1-4 years, n = 65) and low-risk (>4 years, n = 66). The mpMRI-clinical combined model performed better than mpMRI-alone model (AUC:0.84;95 %CI:0.73, 0.92).The stacking model further enhanced CRPC prediction, internal test set (AUC: 0.89; 95 %CI:0.81, 0.93) and external test set (0.82; 95 %CI:0.72, 0.89). In low-risk group, stacking model demonstrated the strongest discriminative capability AUC (0.89;95 %CI:0.77, 0.97).

CONCLUSION: The Stacking model demonstrated favorable predictive capability for CRPC progression risk in advanced Pca, facilitating clinically actionable risk-stratified interventions.

来源:Eur J Radiol,查看原文。版权归原作者所有。

引用

  1. Shen S, Liu B, Li Y, et al. mpMRI-based interpretable machine learning model for predicting castration-resistant prostate cancer risk. Eur J Radiol. 2025. PMID: 41289721. doi:10.1016/j.ejrad.2025.112521

供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。