摘要
目的观察基于平扫CT建立的神经网络深度学习(DL)模型预测保守促排石治疗后排出输尿管结石的价值。方法纳入915例接受保守促排石治疗的输尿管结石患者,随机分为训练集(n=700)、验证集(n=100)及测试集(n=115)。基于平扫CT标记结石三维形状,分别针对训练集和验证集获取三维卷积神经网络(3D-CNN)、二维卷积神经网络(2D-CNN)及全连接神经网络(FCN)最佳参数并建立模型,以测试集检测模型预测能力;绘制受试者工作特征曲线,比较各模型及结石最大径预测测试集经保守治疗后可否排出输尿管结石的效能。结果915例中,229例经保守治疗后排出输尿管结石。3D-CNN模型预测测试集排出输尿管结石的效能最佳,其曲线下面积(AUC)为0.956,高于2D-CNN模型(0.865)、FCN模型(0.813)及结石直径(0.818)(P均<0.01);2D-CNN模型预测AUC高于FCN模型及结石直径(P均<0.05)。结论利用DL模型、尤其3D-CNN能准确预测输尿管结石可否于保守治疗后排出。
Objective To observe the value of deep learning(DL)models established based on plain CT for predicting discharge of ureteral calculus after conservative management.Methods Totally 915 patients with single ureteral calculus who underwent medical expulsive therapy were enrolled.The patients were randomly divided into training set(n=700),validation set(n=100)or test set(n=115).The three-dimensional shape of calculus was marked on plain CT images,and the optimal parameter models of three-dimensional convolutional neural network(3D-CNN),two-dimensional convolutional neural network(2D-CNN)and fully-connected network(FCN)were obtained based on data of training set and verification set.Then receiver operating characteristic curves were drawn,and the efficacies of the models and the maximum diameter of calculus for predicting whether it could be discharged after conservative management were compared.Results Among 915 cases,ureteral calculus was discharged in 229 cases after conservative management.3D-CNN model was the best for predicting whether ureteral calculus could be discharged after conservative management,with the area under the curve(AUC)of 0.956,higher than that of 2D-CNN model(0.865),FCN model(0.813)and calculus diameter(0.818)(all P<0.01).Meanwhile,the AUC of 2D-CNN model was higher than that of FCN model and calculus diameter(both P<0.05).Conclusion DL models,especially 3D-CNN,could be used to accurately predict whether ureteral calculus could be discharged after conservative management.
作者
李金阳
张羽萌
张超
LI Jinyang;ZHANG Yumeng;ZHANG Chao(Department of Urology,Second Hospital of Shandong University,Jinan 250033,China;Department of Medical Imaging,Second Hospital of Shandong University,Jinan 250033,China)
出处
《中国医学影像技术》
CSCD
北大核心
2023年第8期1225-1228,共4页
Chinese Journal of Medical Imaging Technology