基于磁共振表观扩散系数图影像组学的机器学习模型对临床显著性前列腺癌的鉴别诊断价值  

Differential diagnosis of clinically significant prostate cancer using a machine learning model based on magnetic resonance apparent diffusion coefficient imaging

作　　者：林香瑾 朱光斌 张楚若 杜国新 罗锦文[1,2] 关玉宝[1,2] LIN Xiangjin;ZHU Guangbin;ZHANG Churuo;DU Guoxin;LUO Jinwen;GUAN Yubao(The Fifth Clinical School,Guangzhou Medical University,Guangdong Guangzhou 510700,China;Department of Medical Imaging,the Fifth Affiliated Hospital of Guangzhou Medical University,Guangdong Guangzhou 510700,China;The Second Clinical School,Guangzhou Medical University,Guangdong Guangzhou 510700,China)

机构地区：[1]广州医科大学第五临床学院,广东广州510700 [2]广州医科大学附属第五医院医学影像科,广东广州510700 [3]广州医科大学第二临床学院,广东广州510700

出　　处：《现代肿瘤医学》2023年第22期4202-4206,共5页Journal of Modern Oncology

基　　金：广州医科大学科研能力提升计划(编号:02-410-2302087XM);本科生创新能力提升项目(编号:02-408-2304-02XM);广州医科大学第五临床学院本科生创新能力提升计划项目(编号:2022JXA011)。

摘　　要：目的:探讨基于磁共振表观扩散系数图影像组学的机器学习模型对临床显著性前列腺癌的临床应用价值。方法:回顾性分析本院2017年10月至2022年12月经病理证实的182例前列腺癌磁共振检查资料,其中临床显著性前列腺癌(clinically significant prostate cancer,CsPCa)126例,非临床上显著性前列腺癌(clinically insignificant prostate cancer,CiPCa)56例;采用ITK-SNAP软件对磁共振ADC图进行感兴趣区分割,使用pyradiomics软件包,提取包括一阶特征(First Order)、形状特征(Shape)、灰度共生矩阵(GLCM)、灰度区域矩阵(GLSZM)、灰度游程矩阵(GLRLM)、灰度相依矩阵(GLDM)、邻域灰度差分矩阵(NGTDM)影像组学特征;采用一致性检验及最小绝对收缩与选择算子算法(LASSO)筛选出的最佳特征,按7∶3的比例随机分为训练组和验证组,构建逻辑回归(Logistic regression,LR)模型,采用受试者工作特征曲线下面积(area under the curve,AUC)验证其鉴别诊断价值。结果:从ADC图像中提取了1 835个影像组学特征,筛选出15个最佳特征构建逻辑回归机器学习模型。训练集和测试集的准确性分别为0.727、0.700;曲线下面积分别为0.768(95%可信区间:0.700~0.837)、0.719(95%可信区间:0.562~0.875);特异度分别为0.902、0.875;阳性预测值分别为0.683、0.636;阴性预测值分别为0.739、0.718;精确度分别为0.683、0.636。结论:前列腺癌磁共振表观扩散系数图影像组学的逻辑回归机器学习模型,对鉴别临床显著性前列腺癌提供潜在的有效信息。Objective:To investigate the clinical application value of machine learning model based on magnetic resonance apparent diffusion coefficient imaging for clinically significant prostate cancer.Methods:Magnetic resonance examination data of 182 patients with prostate cancer confirmed by pathology from October,2017 to December,2022 in our hospital were retrospectively analyzed,including 126 cases of clinically significant prostate cancer(CsPCa),56 cases of clinically insignificant prostate cancer(CiPCa).ITK-SNAP software was used to segment the area of interest in magnetic resonance ADC image.Pyradiomics software package was used to extract the image omics features including First Order,Shape,GLCM,GLSZM,GLRLM,GLDM and NGTDM.The consistency test,the least absolute shrinkage and selection operator(LASSO) were used to select the best features.These patients were randomly divided into training group and verification group in a ratio of 7:3.Build model of Logistic regression(LR).The area under the curve(AUC) was used to verify its value in differential diagnosis.Results:A total of 1 835 image omics features were extracted from ADC images.15 best features were selected to construct the Logistic regression machine learning model.The accuracy of training set and test set was 0.727 and 0.700 respectively.The area under the curve was 0.768(95% confidence interval:0.700~0.837) and 0.719(95% confidence interval:0.562~0.875) respectively.The specificity was 0.902 and 0.875 respectively.The positive prediction values were 0.683 and 0.636respectively.The negative prediction values were 0.739 and 0.718 respectively.The accuracy was 0.683 and 0.636respectively.Conclusion:The logistic regression machine learning model of magnetic resonance apparent diffusion coefficient imaging of prostate cancer provides potentially effective information for the identification of clinically significant prostate cancer.

关 键 词：临床显著性前列腺癌 非临床上显著性前列腺癌 磁共振成像 表观扩散系数 机器学习模型 

分 类 号：R737.25[医药卫生—肿瘤]