基于AI算法的磁共振检查预约准时与迟到预测的可行性研究及其临床应用  

Feasibility Study of On-Time and Late Prediction of MR Examination Appointments Based on Artificial Intelligence Algorithm and Its Clinical Application

作　　者：任昕 刘水 张晓东[1] 王霄英[1] REN Xin;LIU Shui;ZHANG Xiaodong;WANG Xiaoying(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：《中国医疗设备》2022年第12期44-48,共5页China Medical Devices

基　　金：北京大学第一医院青年临床研究专项基金项目(2018CR25)。

摘　　要：目的探究以人工智能算法为基础建立磁共振检查预约迟到预测模型,对磁共振检查预约患者做出准时与迟到预测的可行性。方法回顾性从放射科信息系统收集2018年1月1日—12月31日连续的已完成磁共振检查的患者预约信息共9087例,经过数据清洗后分为准时组(患者到达时间早于患者预约时间,n=8265),和迟到组(患者到达时间晚于患者预约时间,n=822)。使用XGBoost(Extreme Gradient Boosting)算法作为分类预测模型的基础架构,并结合梯度提升算法中的特征重要性分数生成特征重要性排序列表,训练二分类模型。数据随机分为训练集(70%)和测试集(30%)。以测试集的预测结果检测磁共振检查预约迟到分类预测模型的效能。结果在测试集中,准时者2488例,迟到者239例,磁共振检查预约迟到模型预测准时与迟到的精确度分别为0.994和0.941,召回率分别为0.994和0.933,F1-分数分别为0.994和0.937,ROC曲线下面积均为0.99。结论使用基于XGBoost算法分类模型可对磁共振检查预约做出准时与迟到的预测,为实际临床工作中进一步根据检查项目、检查人数等智能化地分配医学影像科的设备资源、人力资源,以提供更加高效优质的影像检查服务提供了可能。Objective To explore the feasibility of establishing a late prediction model for MR appointment based on artificial intelligence algorithm to predict the punctuality and lateness of MR appointment patients.Methods The appointment data of 9087 consecutive patients who had completed MR examinations from January 1,2018 to December 31,2018 was collected retrospectively from the hospital radiology information system,and after data cleaning,the patients were divided into on time group(patient arrival time earlier than patient appointment time,n=8265)and late group(patient arrival time later than patient appointment time,n=822).The XGBoost(Extreme Gradient Boosting)algorithm was used as the infrastructure for the classification prediction model and combined with the feature importance scores in the gradient boosting algorithm to generate feature importance ranked lists to train the classification models.The data were randomly divided into a training set(70%)and a test set(30%).The prediction results of the test set were used to test the performance of the MR examination appointment lateness classification prediction model.Results In the testing set,2488 cases were punctual and 239 cases were late.The accuracy of the MR appointment lateness model to predict punctuality and lateness was 0.994 and 0.941,the recall rate was 0.994 and 0.933,the F1-score was 0.994 and 0.937,respectively and the area under the ROC curve was 0.99.Conclusion The XGBoost-based classification model is used to predict on-time and late MR appointments,which makes it possible for the medical imaging department to further intelligently allocate the equipment and human resources according to the examination items and the number of examinations in the actual clinical work,so as to provide more efficient and high-quality image examination services.

关 键 词：人工智能 放射科信息系统 磁共振检查预约 准时与迟到 预测 

分 类 号：R445.2[医药卫生—影像医学与核医学] TP391.41[医药卫生—诊断学]