运用影像组学方法鉴别前列腺良恶性病变的诊断价值  

The diagnostic value of radiomics in distinguishing benign and malignant prostatic lesions

作　　者：白伟[1] 张雅楠 郭亚涛 郭炜 刘家祎[3] 徐井旭 黄陈翠 BAI Wei;ZHANG Yanan;GUO Yatao;GUO Wei;LIU Jiayi;XU Jingxu;HUANG Chencui(Department of Radiology,Beijing Jiangong Hospital,Beijing 100054,China;Department of Radiology,Peking University Third Hospital,Beijing 100083,China;Department of Radiology,Beijing Anzhen Hospital,Capital Medical University,Bejing 100029,China;Beiing Shenruibolian Technology Co.Ltd.,Beijing 100080,China)

机构地区：[1]北京市健宫医院影像科,北京100054 [2]北京大学第三医院影像科,北京100083 [3]首都医科大学附属安贞医院影像科,北京100029 [4]北京深睿博联科技有限责任公司,北京100080

出　　处：《实用放射学杂志》2021年第10期1655-1659,共5页Journal of Practical Radiology

摘　　要：目的探讨多参数磁共振成像(mpMRI)的影像组学特征结合机器学习模型在诊断前列腺良恶性病变中的价值.方法在mpMRI图像中提取影像特征并利用LASSO进行特征筛选和降维,采用五折交叉验证来构建逻辑回归(LR)模型用于鉴别前列腺良恶性病变.结果通过LASSO筛选出86个影像特征用于构建诊断LR模型,该模型训练集的准确性为99.14%,敏感性为99.5%,特异性为98.65%,曲线下面积(AUC)值为0.9995.相应的测试集准确性为93.68%,敏感性为95.5%,特异性为91.22%,AUC值为0.9762,2组差异有统计学意义(P<0.05).结论mpMRI的影像组学特征结合机器学习模型诊断前列腺疾病可行性较高,可为临床治疗方案及预后评估提供一种较为可靠的影像诊断依据.Objective To investigate the value of multi parameter magnetic resonance imaging(mpMRI)based radiomics features combined with machine learning model in the diagnosis of benign and malignant prostatic lesions.Methods Image features were extracted from mpMRI images.Feature screening and dimensionality reduction were performed using LASSO,and a logistic regression(LR)model was constructed by using five-fold cross validation to identify benign and malignant prostatic lesions.Results 86 image features were selected by LASSO to construct the LR model.The accuracy,sensitivity,specificity and area under the curve(AUC)of the model were 99.14%,99.5%,98.65% and 0.9995 respectively.The accuracy,sensitivity,specificity and AUC of the corresponding test set were 93.68%,95.5%,91.22% and 0.9762 respectively.The difference between the two groups was statistically significant(P<0.05).Conclusion mpMRI based radiomics features combined with machine learning model is feasible in the diagnosis of prostatic diseases.It can provide reliable imaging diagnosis basis for clinical treatment and prognosis evaluation.

关 键 词：前列腺 良恶性结节 磁共振成像 影像组学 

分 类 号：R737.25[医药卫生—肿瘤] R445.2[医药卫生—临床医学]