摘要： 目的 探讨双参数磁共振成像（MRI）影像组学模型对前列腺超高b值弥散加权成像（DWI）高信号病变的诊断价值。方法 收集中国医科大学北部战区总医院2016年1月至2019年8月经MRI扫描发现前列腺超高b值DWI高信号结节且4周内获得穿刺活检病理结果的81例患者的临床资料，良性34例，恶性47例。患者均行双参数MRI横轴面扫描，参照DWI图像对病灶进行第二版前列腺影像报告和数据系统（PI-RADS v2）评分，利用Dr.Wise多模态科研平台自动勾画DWI （b=3 000 s/mm²）病灶，并以其为标准半自动勾画T2WI压脂及DWI （b=1 000 s/mm²）病灶，对已标注的所有感兴趣区提取影像组学特征，剔除缺失率>10%及自相关性P≥0.9的特征参数，然后用L1正则法降维，再用随机森林法构建诊断模型，以5折交叉验证方法对诊断模型进行检验，测量受试者工作特征（ROC）曲线的曲线下面积（AUC），并以最大约登指数为临界值计算诊断模型的灵敏度和特异度。结果 PI-RADS v2评分鉴别前列腺良恶性病变ROC的AUC为0.653，灵敏度及特异度分别为0.717、0.486。对81例患者靶病灶提取得到的特征进行相关性分析、降维及随机森林法构建诊断模型后，T2WI压脂、DWI （b=1 000 s/mm²、b=3 000 s/mm²）分别得到具有原始信息的特征9、10、9个，其中包括2个共有特征（伸长率、球形度），3个研究序列区分验证集前列腺病变良恶性的AUC分别为0.723、0.800和0.833；灵敏度分别为0.696、0.732、0.821；特异度分别为0.689、0.733、0.756。结论 双参数MRI影像组学模型对前列腺超高b值DWI高信号病变良恶性具有鉴别诊断价值，诊断效能高于PI-RADS v2评分系统。

双参数MRI影像组学模型对前列腺超高b值DWI高信号病变的诊断价值

王聪¹, 刘娜¹, 包丝雨², 曹乐¹, 辛迪¹, 卑贵光¹

1. 中国医科大学北部战区总医院放射诊断科, 沈阳 110812;
2. 北京深睿博联科技有限责任公司研发中心, 北京 100080

收稿日期:2020-09-01出版日期:2021-10-30发布日期:2021-10-11
通讯作者:卑贵光E-mail:2023477349@qq.com
作者简介:王聪(1991-),女,主治医师,硕士研究生.
基金资助:辽宁省自然科学基金重点项目（20170540916）


关键词: 前列腺肿瘤, 磁共振成像, 影像组学, 诊断价值
Abstract: Objective To explore the diagnostic value of the biparametric magnetic resonance imaging (MRI) model in the diagnosis of hyperintensity diffusion weighted imaging (DWI) lesions of the prostate with an ultra-high b value. Methods Clinical data were collected from 81 patients diagnosed with hyperintensity DWI lesions of the prostate with an ultra-high b value on MRI whose biopsy and pathology results were obtained within 4 weeks from January 2016 to August 2019 at Northern Theater General Hospital, China Medical University. Among them, 34 patients had benign lesions and 47 patients had malignant lesions. All patients underwent biparametric MRI in the transverse plane. The second edition of the prostate imaging of reporting and data system (PI-RADS v2) was used to score the lesions according to DWI images. Draw DWI (b=3 000 s/mm²) lesions were automatically delineated using the Beijing Dr.Wise multimodal scientific research platform, and T2-weighted fat pressing imaging and DWI (b=1 000 s/mm²) lesions were semi-automatically delineated using it as the standard. The image group features were extracted from all the labeled regions of interest, and the feature parameters with a deletion rate of >10% and an autocorrelation P ≥ 0.9 were eliminated; then, an L1-based method was used to reduce dimensions, and the random forest (RF) method was used to build a diagnostic model, which was tested using the five-fold cross-validation method. The area under the curve (AUC) of the receiver operating characteristic (ROC) curve was measured, and the sensitivity and specificity of the diagnostic model were calculated using the maximum approximate index as the critical value. Results The AUC of the PI-RADS v2 score was 0.653, and the sensitivity and specificity were 0.717 and 0.486, respectively. After correlation analysis, dimensionality reduction, and using the RF method, 9, 10, and 9 features with original information were obtained by T2-weighted fat pressing imaging, DWI with b=1 000 s/mm², and DWI with b=3 000 s/mm², respectively, including 2 common features (elongation and sphericity). The AUCs of benign and malignant prostate lesions for T2-weighted fat pressing imaging, DWI with b=1 000 s/mm², and DWI with b=3 000 s/mm² were 0.723, 0.800, and 0.833, respectively. The sensitivity was 0.696, 0.732, and 0.821, respectively, while the specificity was 0.689, 0.733 and 0.756, respectively. Conclusion The radiomics model based on biparametric MRI is valuable in differentiating benign lesions from malignant lesions using the high b-value DWI signal, and its diagnostic efficiency is higher than that of the second edition of the PI-RADS v2.
Key words: prostate neoplasms, magnetic resonance imaging, radiomics, diagnostic value
PDF全文下载地址:

https://journal.cmu.edu.cn/CN/article/downloadArticleFile.do?attachType=PDF&id=2858