曾小辉,彭涛,高月琴,牛翔科,陈雪卉,张仕慧,陈志凡.基于前列腺影像报告和数据系统第2版的机器学习模型诊断高级别前列腺癌[J].中国医学影像技术,2018,34(12):1852~1856
基于前列腺影像报告和数据系统第2版的机器学习模型诊断高级别前列腺癌
Evaluation of high-grade prostate cancer with machine learning model of prostate imaging reporting and data system version 2
投稿时间:2018-08-30  修订日期:2018-11-05
DOI:10.13929/j.1003-3289.201808198
中文关键词:  前列腺肿瘤  支持向量机  决策树  Logistic模型
英文关键词:Prostatic neoplasms  Support vector machine  Decision tree  Logistic models
基金项目:四川省卫计委科研项目(18PJ150)、成都市卫计委科研项目(2015080)。
作者单位E-mail
曾小辉 成都大学附属医院放射科, 四川 成都 610081  
彭涛 成都大学附属医院放射科, 四川 成都 610081 474488159@qq.com 
高月琴 成都大学附属医院放射科, 四川 成都 610081  
牛翔科 成都大学附属医院放射科, 四川 成都 610081  
陈雪卉 成都大学附属医院放射科, 四川 成都 610081  
张仕慧 成都大学附属医院放射科, 四川 成都 610081  
陈志凡 成都大学附属医院放射科, 四川 成都 610081  
摘要点击次数: 1657
全文下载次数: 885
中文摘要:
      目的 建立基于前列腺影像报告和数据系统第2版(PI-RADS v2)的支持向量机(SVM)、决策树(DT)和Logistic回归3种机器学习模型,评价上述模型对高级别前列腺癌的诊断价值。方法 回顾性分析于我院接受前列腺多参数MR扫描并取得病理结果的194例患者的资料,其中高级别癌63例,非高级别癌131例。将评价因素(PI-RADS v2评分、年龄、游离前列腺特异抗原、前列腺特异性抗原比值、前列腺特异抗原密度)录入SVM、DT和Logistic回归3种机器学习模型进行诊断,通过ROC曲线评价PI-RADS v2评分和3种机器学习模型诊断高级别前列腺癌的价值。结果 PI-RADS v2、SVM、DT和Logistic回归模型诊断高级别前列腺癌的敏感度分别为72.73%、69.09%、87.27%和70.91%;特异度分别为87.29%、93.22%、93.22%和95.76%。DT模型诊断高级别前列腺癌ROC的AUC(AUC=0.90,P<0.01)最大,且与PI-RADS v2评分、SVM、Logistic回归比较差异均有统计学意义(P均<0.05)。结论 PI-RADS v2评分、SVM、DT和Logistic回归模型诊断高级别前列腺癌的价值均较好。
英文摘要:
      Objective To establish the support vector machine (SVM), decision tree (DT) and Logistic regression models with prostate imaging reporting and data system version 2 (PI-RADS v2), and to evaluate their diagnostic efficiency in high-grade prostate cancer. Methods Clinical and MRI data of histopathologically proved prostate cancer in 194 patients were retrospectively analyzed, including 63 patients of high-grade cancer group and 131 patients of non high-grade cancer group. The evaluation factors, including PI-RADS v2 score, age, free prostate specific antigen, prostate specific antigen ratio and adjusted-prostate-specific antigen density were analyzed with SVM, DT and Logistic models, and the values of PI-RADS v2 and three models in diagnosis of high-grade prostate cancer were evaluated with ROC. Results The sensitivity of PI-RADS v2, SVM, DT and Logistic models in diagnosis of high-grade prostate cancer was 72.73%, 69.09%, 87.27% and 70.91%, while the specificity was 87.29%, 93.22%, 93.22% and 95.76%, respectively. AUC of DT model in diagnosis of high-grade prostate cancer was the largest (AUC=0.90, P<0.01). There were statistically significant differences of AUC of DT and the PI-RADS v2 score, SVM, and Logistic models (all P<0.05). Conclusion PI-RADS v2, SVM, DT and Logistic models has good value in diagnosis of high-grade prostate cancer.
查看全文  查看/发表评论  下载PDF阅读器