基于深度学习的早期急性胰腺炎严重程度分级预测  

Early acute pancreatitis severity grading prediction based on deep learning

作　　者：邓鸿 郭世尧 冯文 李文雪 肖波[2] 何汶静 DENG Hong;GUO Shi-yao;FENG Wen(School of Medical Imaging,North Sichuan Medical College,Sichuan 637000,China)

机构地区：[1]川北医学院医学影像学院,四川南充637000 [2]重庆医科大学附属璧山医院(重庆市璧山区人民医院),重庆402760

出　　处：《放射学实践》2025年第3期369-376,共8页Radiologic Practice

基　　金：国家自然科学基金专项项目(62341501);南充市市校科技战略合作项目(22SXQT0323);南充市市校科技战略合作项目(22SXQT0325)。

摘　　要：目的:建立基于急性胰腺炎(AP)患者早期静脉期增强CT图像对AP进行严重程度分类预测的深度学习(DL)模型。方法:回顾性分析川北医学院附属医院收治的215例首发AP患者静脉期增强CT资料,按照采集时间的先后,将数据影像资料划分为训练和验证集(141例,其中男87例,女54例;平均年龄51.37±16.09岁)以及测试集(74例,其中男41例,女33例;平均年龄55.49±17.83岁),利用卷积神经网络建立早期AP分类器模型,使用5折交叉验证策略训练验证模型效能。以受试者操作特征(ROC)曲线下面积(AUC)、准确率、敏感度、特异度、F1分数来评估模型的预测效能。结果:早期AP分类器模型表现出良好的预测效能,在训练集、验证集中的AUC分别为1.000、1.000,在测试集中的AUC为0.806、敏感度为0.850、特异度为0.675、F1分数为0.730、准确率为0.770。结论:基于静脉期增强CT图像的深度学习分类模型预测AP重症患者的效果良好,可以为AP的临床诊断预测提供辅助性决策。Objective:Establish a deep learning(DL)model for classification and prediction of the severity of acute pancreatitis(AP)based on early venous phase enhanced computed tomography(CT)images of patients with AP.Methods:A retrospective analysis was conducted on venous phase enhanced CT data from 215 first-time AP patients admitted to the Affiliated Hospital of North Sichuan Medical College,according to the order of collection time,the data were divided into training and validation set(141 cases,including 87 males and 54 females;average age 51.37±16.09 years)and testing set(74 cases,including 41 males and 33 females;average age 55.49±17.83 years).An early AP classifier model was developed using Convolutional Neural Networks(CNN),trained and validated using a 5-fold cross-validation strategy.The predictive performance of the model was evaluated based on the Area Under the Curve(AUC)of the Receiver Operating Characteristic(ROC),accuracy,sensitivity,specificity,and F1 score.Results:Early AP classifier models showed good predictive performance,The AUC values in the training set,validation set,and test set were 1.000,1.000,and 0.806,respectively;In the test set,the sensitivity was 0.850,the specificity was 0.675,the F1 score was 0.730,and the accuracy was 0.770.Conclusion:The deep learning classification model based on venous phase enhanced CT images effectively predicts severity AP patients and can assist in clinical diagnosis and decision-making.

关 键 词：急性胰腺炎 深度学习 体层摄影术 X线计算机 

分 类 号：R814.42[医药卫生—影像医学与核医学] R576[医药卫生—放射医学]