基于MRI影像组学对前列腺癌根治术后Gleason评分升级的预测研究  

Predictive study on Gleason score upgrading after radical prostatectomy for prostate cancer based on MRI radiomics

作　　者：李志平 崔凤[1] 张永胜[1] 徐辉景 杨丽勤 杜亮 李焕国 石徐 张育 诸靖宇[3] LI Zhiping;CUI Feng;ZHANG Yongsheng;XU Huijing;YANG Liqin;DU Liang;LI Huanguo;SHI Xu;ZHANG Yu;ZHU Jingyu(Department of Radiology,Hangzhou TCM Hospital Affiliated to Zhejiang Chinese Medical University,Hangzhou 310007,China;不详)

机构地区：[1]浙江中医药大学附属杭州市中医院放射科,310007 [2]浙江中医药大学附属杭州市中医院病理科,310007 [3]浙江中医药大学附属杭州市中医院泌尿外科,310007

出　　处：《浙江医学》2025年第4期356-361,391,I0004,I0005,共9页Zhejiang Medical Journal

基　　金：浙江省中医药科技计划项目(2024ZL688、2024ZL668);浙江省医药卫生科技计划项目(2022KY996、2024KY1386);杭州市农业和社会发展项目(20231203A12);杭州市卫生科技计划项目(A20230086)。

摘　　要：目的探讨基于MRI影像组学特征预测前列腺癌(PCa)根治术后Gleason评分(GS)升级的价值。方法回顾性分析2015年1月至2024年6月浙江中医药大学附属杭州市中医院经病理检查证实的PCa患者175例。其中,2022年6月1日之前的120例PCa患者作为训练队列,6月1日之后的55例PCa患者作为验证队列。使用PyRadiomics软件从表观扩散系数图中提取影像组学特征。采用最小绝对收缩和选择算子算法筛选关键特征,构建优化后的特征集。通过十倍交叉验证法确定最佳特征组合,并计算各特征的权重。结合临床指标、影像学特征建立预测GS升级的联合模型,通过准确度、灵敏度、特异度、F1分数、ROC曲线、决策曲线和校准曲线评估其效能。结果训练队列120例中GS未升级77例,GS升级43例;验证队列55例中GS未升级28例,GS升级27例。单因素和多因素分析显示初始GS和年龄是预测GS升级的独立危险因素(均P<0.05)。基于影像组学、初始GS和年龄3个指标构建预测GS升级的联合模型。在训练队列中,该联合模型的AUC为0.902,高于单独临床和影像组学模型的AUC。在验证队列中,该联合模型的AUC为0.804,高于单独临床和影像组学模型的AUC。决策曲线显示该联合模型在预测概率方面较好,校准曲线显示预测概率和实际概率之间具有较好的一致性。结论基于MRI影像组学对PCa患者根治性前列腺切除术后GS升级具有较高的预测价值,与临床指标联合使用可提高其预测效能。Objective To explore the value of preoperative MRI radiomic features in predicting Gleason score(GS)upgrading after radical prostatectomy for prostate carcinoma(PCa).Methods A retrospective analysis was conducted on 175 PCa patients confirmed by pathology at Hangzhou TCM Hospital Affiliated to Zhejiang Chinese Medical University from January 2015 to June 2024.Among them,120 PCa patients collected before June 1st,2022 were used as the training cohort,while another 55 PCa patients collected after June 1st,2022 were used as the validation cohort.Radiomic features were extracted from apparent diffusion coefficient maps using PyRadiomics software.Key features were selected using the least absolute shrinkage and selection operator algorithm,and an optimized feature set was constructed.The best feature combination was determined by ten-fold cross-validation,and the weights of each feature were calculated.A predictive model for GS upgrading was established by combining clinical indicators and imaging features,and the model performance was evaluated via accuracy,sensitivity,specificity,F1 score,ROC curve,decision curve,and calibration curve.Results Among the 120 patients in the training cohort,77 had no GS upgrading while 43 had GS upgrading;of 55 patients in the validation cohort,28 had no GS upgrading while 27 had GS upgrading.Univariate and multivariate analyses showed that initial GS and age were independent risk factors for GS upgrading(both P<0.05).A combined model for GS upgrading was constructed based on radiomics,initial GS,and age.In the training cohort,the AUC of the combined model was 0.902,higher than that of the clinical and radiomic models alone.In the validation cohort,the AUC of the model was 0.804,higher than that of the clinical and radiomic models alone.The decision curve demonstrated favorable benefits of the combined model on patient intervention based on predicted probabilities,and the calibration curve showed good consistency between predicted and actual probabilities.Conclusion MRI radiomics has a

关 键 词：前列腺癌 Gleason评分升级 MRI 影像组学 预测模型 

分 类 号：R73[医药卫生—肿瘤]