基于双参数MRI的影像组学特征融合方法预测前列腺腺癌Gleason分级的研究

Study of Radiomics Features Fusion Based on Biparametric-MRI in Predicting Gleason Grade of Prostate Adenocarcinoma

目的探讨基于磁共振(Magnetic Resonance Imaging,MRI)双参数影像组学特征融合方法对前列腺腺癌(Prostate Adenocarcinoma,PA)的Gleason分级的诊断价值。方法回顾性收集蚌埠医学院第一附属医院和蚌埠医学院第二附属医院经病理证实为PA的患者158例,其中低级别组(Gleason评分≤3+4)89例,高级别组(Gleason评分≥4+3)69例。所有患者获取病理前均经过MRI检查,由两组医生分别对每位患者的T_(2)WI和表观弥散系数(Apparent Diffusion Coefficient,ADC)图像进行感兴趣区(Region of Interest,ROI)分割,再经过影像组学进行特征提取、筛选,建立模型。运用ROC曲线及曲线下面积(Area Under Curve,AUC)分别评估T_(2)WI、ADC、T_(2)WI+ADC融合模型的诊断效能。结果按照7∶3的比例将患者随机分层分配到训练集(110例)和测试集(48例)。通过训练集的特征筛选,T_(2)WI、ADC、T_(2)WI+ADC分别筛选出10、5、7个特征,并建立Logistic模型用来鉴别低级别组与高级别组PA。训练集中各模型鉴别诊断的AUC值分别为0.768(95%CI:0.681~0.856)、0.759(95%CI:0.671~0.847)、0.835(95%CI:0.759~0.911);测试集中各模型鉴别诊断的AUC值分别为0.638(95%CI:0.466~0.811)、0.700(95%CI:0.545~0.845)、0.808(95%CI:0.681~0.935)。结论基于多中心双参数MRI构建的影像组学模型具备鉴别PA Gleason低级别组与高级别组的能力,并且T_(2)WI+ADC融合模型具有较高的预测价值,可为临床分级提供参考。

Objective To explore the diagnostic value of Gleason grading of prostate adenocarcinoma(PA)based on biparametricmagnetic resonance imaging(MRI)radiomics features fusion.Methods A total of 158 male patients with PA confirmed by pathology in the First Affiliated Hospital of Bengbu Medical College and the Second Affiliated Hospital of Bengbu Medical College were collected retrospectively,including 89 cases in the low-grade group(Gleason score≤3+4)and 69 cases in the high-grade group(Gleason score≥4+3).All patients were examined by MRI before obtaining pathology.The T_(2)WI and apparent diffusion coefficient(ADC)images of each patient were manually segmented by layer by two groups of doctors.The segmented regions of interest were extracted and screened by radiomics features to establish models.Receiver operating characteristic curve and area under the curve(AUC)were used to evaluate the diagnostic efficiency of T_(2)WI,ADC and T_(2)WI+ADC fusion models respectively.Results All patients were randomly assigned to training set(110 cases)and test set(48 cases)according to the ratio of 7∶3.Through the feature screening of the training set,T_(2)WI,ADC,T_(2)WI+ADC screened out 10,5,7 features respectively,and a Logistic model was established to identify low-grade and high-grade PA.The AUC values of model in the training set were 0.768(95%CI:0.681-0.856),0.759(95%CI:0.671-0.847)and 0.835(95%CI:0.759-0.911)respectively;the AUC values of model in the test set were 0.638(95%CI:0.466-0.811),0.700(95%CI:0.545-0.845)and 0.808(95%CI:0.681-0.935)respectively.Conclusion The radiomics model based on multi-center bp-MRI can identify low-grade Gleason group from high-grade Gleason group in PA,and the T_(2)WI+ADC fusion model has high predictive value,which can provide reference for clinical grading.