基于多模态数据构建前列腺癌诊断预测模型  

Building a diagnosis and prediction model for prostate cancer based on multimodal data

作　　者：沈登文 蓝思荣[1] 李雄 陈南辉[2] 张添辉 江惠明 Shen Dengwen;Lan Sirong;Li Xiong;Chen Nanhui;Zhang Tianhui;Jiang Huiming(Department of Ultrasound,Meizhou People′s Hospital,Meizhou 514031,China;Department of Urology,Meizhou People′s Hospital,Meizhou 514031,China;Department of Radiology,Meizhou People′s Hospital,Meizhou 514031,China)

机构地区：[1]梅州市人民医院超声科,梅州514031 [2]梅州市人民医院泌尿外科,梅州514031 [3]梅州市人民医院磁共振科,梅州514031

出　　处：《中国医师杂志》2023年第8期1139-1143,共5页Journal of Chinese Physician

基　　金：梅州市社会发展科技计划项目(2022C0301035)。

摘　　要：目的探讨基于临床、多参数磁共振成像联合经直肠超声弹性数据对前列腺癌的诊断价值。方法回顾性分析2021年11月至2023年3月期间在梅州市人民医院同时进行了经直肠前列腺二维超声、前列腺实时应变弹性成像、前列腺计算机断层扫描和多参数磁共振成像(MP-MRI)检查及前列腺穿刺活检的患者资料。收集患者年龄、身高、体重、游离血清前列腺特异性抗原(fPSA)、总前列腺特异性抗原(tPSA)、fPSA/tPSA、MRI前列腺影像报告和数据系统(PI-RADS)评分和超声弹性值。用多因素logistics回归构建四个前列腺癌诊断预测模型进行对比,选取最优模型构建列线图。采用受试者工作特征(ROC)曲线评价不同模型的诊断性能,采用校准曲线评价列线图的诊断性能。结果共纳入117例患者117个前列腺病灶,47个良性前列腺病变和70个前列腺癌病灶,良恶性病灶患者年龄、fPS A、tPSA、fPSA/tPSA、PI-RADS评分以及超声弹性值差异有统计学意义(均P<0.01)。临床模型(年龄+tPSA+fPSA+fPSA/tPSA)、MRI模型(PI-RADS评分)、超声弹性模型、临床+MRI+超声弹性联合模型诊断前列腺癌的曲线下面积(AUC)分别为0.86、0.86、0.92、0.98。结论与单一诊断模型相比,年龄、tPSA、fPSA/tPSA、PI-RADS评分联合超声弹性值模型能提高前列腺癌诊断率。Objective To explore the diagnostic value of clinical,multi-parameter magnetic resonance imaging(MP-MRI)combined with transrectal ultrasound elasticity data for prostate cancer.Methods A retrospective analysis was conducted on patient data from November 2021 to March 2023 when transrectal prostate two-dimensional ultrasound,real-time strain elastography of the prostate,MP-MRI examination of the prostate,and prostate biopsy were performed simultaneously at the Meizhou People′s Hospital.We collected patient age,height,weight,free serum prostate specific antigen(fPSA),total prostate specific antigen(tPSA),fPSA/tPSA,MRI prostate imaging report and data system(PI-RADS)scores,and ultrasound elasticity values.Four predictive models for prostate cancer diagnosis were constructed using multivariate logistic regression for comparison,and the optimal model was selected to construct a column chart.The diagnostic performance of different models was evaluated using receiver operating characteristic(ROC)curves,and the diagnostic performance of column charts was evaluated using calibration curves.Results This study included a total of 117 patients with 117 prostate lesions,47 benign prostate lesions,and 70 prostate cancer lesions.There were statistically significant differences in age,fPSA,tPSA,fPSA/tPSA,PI-RADS scores,and ultrasound elasticity values between benign and malignant lesions patients(all P<0.01).The area under the curve(AUC)of the clinical model(age+tPSA+fPSA+fPSA/tPSA),MRI model(PI-RADS score),ultrasound elastic model,and clinical+MRI+ultrasound elastic combined model for diagnosing prostate cancer were 0.86,0.86,0.92,and 0.98,respectively.Conclusions Compared with a single diagnostic model,the combination of age,tPSA,fPSA/tPSA,PI-RADS scores,and ultrasound elasticity value model can improve the diagnostic rate of prostate cancer.

关 键 词：前列腺肿瘤 弹性成像技术 多参数磁共振成像 模型 统计学 

分 类 号：R737.25[医药卫生—肿瘤]