摘要
目的比较基于双参数MRI的深度学习模型和临床模型对临床显著性前列腺癌(clinical significant prostate cancer,csPCa)的诊断价值,并联合深度学习模型和临床指标建立混合模型,探讨混合模型对csPCa诊断效能的提升价值。材料与方法回顾性分析本院2017年2月至2022年5月共531例因临床怀疑前列腺癌而行术前MRI并行后续穿刺和/或手术病理检查患者的临床及影像资料,其中csPCa 319例,非csPCa 212例。按照8∶2比例随机划分为训练集(425例)和测试集(106例)。手动勾画T2加权成像(T2-weighted imaging,T2WI)、扩散加权成像(diffusion-weighted imaging,DWI)及表观扩散系数(apparentdiffusion coefficient,ADC)图像的感兴趣区后采用DenseNet网络建立深度学习模型,采用单因素和多因素逻辑回归筛选出临床特征后建立临床模型,并使用逻辑回归联合深度学习模型和临床特征建立深度学习-临床混合模型。使用受试者工作特征(receiver operating characteristic,ROC)曲线评估模型性能,使用DeLong检验比较曲线下面积(area under the curve,AUC)。结果逻辑回归分析显示年龄、前列腺特异性抗原(prostate specific antigen,PSA)及前列腺影像报告和数据系统(prostate imaging reporting and data system,PI-RADS)评分为csPCa的独立危险因素。在测试集中,深度学习模型的AUC值为0.90[95%置信区间(confidence interval,CI):0.85~0.96],临床模型的AUC值为0.85(95%CI:0.78~0.92),两者间差异无统计学意义(P=0.245)。深度学习-临床混合模型的AUC值为0.93(95%CI:0.88~0.98),优于临床模型(P=0.034)和深度学习模型(P=0.048)。结论深度学习模型对csPCa的诊断效能与临床模型相当;深度学习-临床混合模型对csPCa的诊断效能最高,具有良好的应用价值,可作为临床诊断csPCa的辅助工具。
Objective:To compare the diagnostic performance of the deep learning model based on bi-parametric MRI with a clinical model for clinically significant prostate cancer(csPCa)and explore the value of a combined model incorporating deep learning model and clinical variables to enhance the diagnostic efficacy of csPCa.Materials and Methods:Imaging and clinical data from 531 patients(319 csPCa and 212 non-csPCa)who underwent pre-operative MRI and subsequent biopsy and/or surgical pathology examination for clinically suspected PCa at our hospital from February 2017 to May 2022 were retrospectively analyzed.The patients were randomly divided into a training cohort(425 cases)and a testing cohort(106 cases)at a ratio of 8∶2.The volumes of interests were manually segmented on T2-weighted imaging(T2WI),diffusion-weighted imaging(DWI),and its derivative apparent diffusion coefficient(ADC)maps and a deep learning model was developed utilizing the DenseNet network.Through univariate and multivariate logistic regressions,clinical features were selected to build a clinical model.A deep learning-clinical combined model was created by integrating the output of the deep learning model with clinical variables based on logistic regression.The receiver operating characteristic(ROC)curve was used to assess the model performance,and the DeLong test was employed to compare the diagnostic performance of different models.Results:Logistic analyses showed that age,prostate specific antigen(PSA)value and prostate imaging reporting and data system(PI-RADS)score were significant factors for predicting csPCa.In the testing set,the AUC of the deep learning model was 0.90[95%confidence interval(CI):0.85-0.96],which showed no significant difference with the clinical model[0.85(95%CI:0.78-0.92),P=0.245].The AUC of the deep learning-clinical combined model reached 0.93(95%CI:0.88-0.98),which significantly outperformed both the clinical model(P=0.034)and the deep learning model(P=0.048).Conclusions:The diagnostic performance of the deep learning model for csPCa was comparable to the clinical model.The deep learning-clinical combined mode achieved the highest diagnostic efficacy,which possessed good practical utility and could be utilized as an auxiliary method for clinical diagnosis of csPCa.
作者
胡尘翰
乔晓梦
胡冀苏
包婕
曹昌浩
王希明
HU Chenhan;QIAO Xiaomeng;HU Jisu;BAO Jie;CAO Changhao;WANG Ximing(Department of Radiology,First Affiliated Hospital of Soochow University,Suzhou 215006,China;Suzhou Institute of Biomedical Engineering and Technology,Chinese Academy of Sciences,Suzhou 215006,China)
出处
《磁共振成像》
CAS
CSCD
北大核心
2024年第2期90-96,共7页
Chinese Journal of Magnetic Resonance Imaging
基金
苏州市医疗卫生科技创新项目(编号:SKY2022003)
苏州市临床重点病种诊疗技术专项项目(编号:LCZX202001)
苏州市科教兴卫青年科技项目(编号:KJXW2023006)。
关键词
前列腺癌
磁共振成像
深度学习
机器学习
诊断效能
prostate cancer
magnetic resonance imaging
deep learning
machine learning
diagnostic efficacy
