引用本文
  • 邢朋毅,宋涛,王铁功,陈录广,马超,阳青松,陆建平*.基于双参数磁共振成像影像组学机器学习的前列腺癌风险分层[J].第二军医大学学报,2021,42(3):233-242    [点击复制]
  • XING Peng-yi,SONG Tao,WANG Tie-gong,CHEN Lu-guang,MA Chao,YANG Qing-song,LU Jian-ping*.Risk stratification of prostate cancer based on biparametric magnetic resonance imaging radiomics machine learning[J].Acad J Sec Mil Med Univ,2021,42(3):233-242   [点击复制]
【打印本页】 【下载PDF全文】 【HTML】 查看/发表评论下载PDF阅读器关闭

←前一篇|后一篇→

过刊浏览    高级检索

本文已被:浏览 320次   下载 256 本文二维码信息
码上扫一扫!
基于双参数磁共振成像影像组学机器学习的前列腺癌风险分层
邢朋毅,宋涛,王铁功,陈录广,马超,阳青松,陆建平*
0
(海军军医大学(第二军医大学)长海医院影像医学科, 上海 200433
*通信作者)
摘要:
目的 探讨基于双参数MRI影像组学特征的机器学习模型在前列腺癌风险分层中的作用。方法 收集128例经病理证实的前列腺癌患者的临床资料,其中低风险组(Gleason评分 ≤ 3+4分)60例、高风险组(Gleason评分 ≥ 4+3分)68例。所有患者均接受3.0 T MRI检查,采集参数相同。统计与前列腺癌相关的临床危险因素,包括年龄、病灶体积、病灶位置、前列腺特异性抗原及前列腺影像报告和数据系统(PI-RADS)评分等。按7:3将患者随机分为训练集和验证集,分别用于影像组学模型的机器学习和验证。影像组学特征包括基于梯度的直方图特征、形态特征、灰度共生矩阵(GLCM)、灰度游程矩阵(GLRLM)、灰度大小区域矩阵(GLSZM)和Haralick特征。应用多因素logistic回归分析建立3个前列腺癌风险分层的预测模型:临床模型、影像组学模型和临床-影像组学联合模型,分别通过ROC曲线和决策曲线分析比较各模型的诊断效能与临床效益。结果 影像组学模型和临床-影像组学联合模型对验证集的预测效能相当(AUC=0.78,95% CI 0.63~0.93),并且均优于临床模型(AUC=0.75,95% CI 0.60~0.91)。决策曲线分析表明,影像组学模型和临床-影像组学联合模型比临床模型具有更高的临床净收益。结论 与仅评估前列腺癌相关的临床危险因素相比,基于双参数MRI影像组学的临床-影像组学机器学习模型可以提高对前列腺癌风险分层预测的准确性。
关键词:  前列腺肿瘤  磁共振成像  影像组学  机器学习  危险性评估
DOI:10.16781/j.0258-879x.2021.03.0233
投稿时间:2020-08-28修订日期:2020-12-30
基金项目:国家临床重点专科军队建设项目(总后卫生部),上海市卫生与计划生育委员会面上项目(M20140149).
Risk stratification of prostate cancer based on biparametric magnetic resonance imaging radiomics machine learning
XING Peng-yi,SONG Tao,WANG Tie-gong,CHEN Lu-guang,MA Chao,YANG Qing-song,LU Jian-ping*
(Department of Radiology, Changhai Hospital, Naval Medical University (Second Military Medical University), Shanghai 200433, China
*Corresponding author)
Abstract:
Objective To explore the role of radiomics machine learning model based on biparametric magnetic resonance imaging (MRI) in the risk stratification of prostate cancer. Methods The clinical data of 128 patients with histologically proven prostate cancer were collected, including 60 cases in low-risk group (Gleason score ≤ 3+4) and 68 cases in high-risk group (Gleason score ≥ 4+3). All the patients were examined by 3.0 T MRI with the same parameters, and the clinical risk factors related to prostate cancer (age, volume of prostate lesions, location of lesions, prostate-specific antigen and prostate imaging reporting and data system[PI-RADS]score) were analyzed. The patients were randomly assigned (7:3) to training set or validation set for radiomics machine learning and verification. The radiomics features included gradient-based histogram features, morphological features, gray-level co-occurrence matrix (GLCM), gray-level run-length matrix (GLRLM), gray-level size zone matrix (GLSZM) and Haralick features. Multivariate logistic regression analysis was used to establish 3 prediction models to stratify the risks of prostate cancer:clinical model, radiomics model and clinical-radiomics combined model. The diagnostic performance and clinical benefits of each model were compared by receiver operating characteristic (ROC) curve and decision curve. Results The predictive efficacy of the radiomics model and the clinical-radiomics combined model in validation set were similar (area under curve[AUC]=0.78, 95% confidence interval[CI]0.63-0.93) and were better than that of the clinical model (AUC=0.75, 95% CI 0.60-0.91). Decision curve analysis showed that the radiomics model and the clinical-radiomics model had higher clinical net benefits than the clinical model. Conclusion Compared with only evaluating the clinical risk factors related to prostate cancer, the clinical-radiomics machine learning model based on biparametric MRI radiomics can improve the predictive accuracy of risk stratification of prostate cancer.
Key words:  prostatic neoplasms  magnetic resonance imaging  radiomics  machine learning  risk assessment