基于深度学习算法的增强CT检查后对比剂肾病预测模型研究  

A preliminary study on the prediction model of contrast-medium nephropathy after contrast-enhanced CT based on deep learning algorithm

作　　者：赵凯[1] 吴静云[1] 张保翠[1] 罗健[1] 张晓东[1] 王霄英[1] ZHAO Kai;WU Jing-yun;ZHANG Bao-cui(Department of Radiology,Peking University First Hospital,Beijing 100034,China)

机构地区：[1]北京大学第一医院医学影像科,北京100034

出　　处：《放射学实践》2023年第6期720-725,共6页Radiologic Practice

摘　　要：目的:利用深度学习算法对增强CT检查后发生对比剂肾病(CIN)的风险因素进行分析,并构建CIN的预测模型。方法:从RIS系统中回顾性搜集增强CT检查并建立CIN数据库。检索数据库资料,导出基本信息、基础病史、对比剂注射信息共计18项指标,对患者资料进行筛选、预处理后,建立二分类模型研究队列。经数据处理后利用深度学习方法在整理好的CIN数据集上进行开发和训练。结果:CIN二分类模型对测试集数据预测结果显示CIN阴性分类的精确度、召回率和F1-分数分别为0.982、0.752和0.852,CIN阳性分类的精确度、召回率和F1-分数分别为0.229、0.842和0.359。该模型ROC曲线下面积均为0.89。结论:本研究使用深度学习算法构建了CIN的预测模型,模型对CIN阳性的患者有较高敏感性,但是特异性有待提高。Objective:To analyze the risk factors of contrast-induced nephropathy(CIN)after contrast-enhanced CT examination and to construct a model for CIN prediction by deep learning algorithm.Methods:The data of contrast-enhanced CT examinations were retrospectively collected from the RIS system in the hospital,and a CIN database was established.A total of 18 indicators,including basic information,basic medical history,and contrast agent injection information,were derived from the database.After data screening and preprocessing,a two-category model study cohort was established.After data processing,deep learning methods were used to develop and train on the sorted CIN dataset.Results:The prediction results of the test data set showed that the precision,recall,and F1-score of CIN-negative classification were 0.982,0.752,and 0.852,respectively.The precision,recall,and F1-score of CIN-positive classification were 0.229,0.842,and 0.359,respectively.The area under the ROC curve of the model was 0.89.Conclusion:This study constructs a CIN prediction model based on a deep learning algorithm,which has high sensitivity for CIN-positive patients,but the specificity needs to be improved.

关 键 词：深度学习 人工智能 体层摄影术 X线计算机 肾病 危险因素 

分 类 号：R-056[医药卫生] R-05R814.4R692R181.13