摘要
冠心病早期发现可显著提高治愈率,但受限于其前驱症状不明显、患者自我认识不足等原因,冠心病早期发现率较低。通过构建基于电子病历时间嵌入的编码器稠密网络冠心病预测模型E-TEED-CHD,可以有效使用居民从出生起跨区域、跨机构、跨病种的历次非结构化电子病历数据进行冠心病早期预测。该模型将居民历次电子病历分别嵌入时间位信息,并行输入多个编码器进行加速处理;采用稠密网络模型对编码器输出进行解析,实现冠心病的早期预测。进一步地,E-TEED-CHD模型使用一种新颖的病历掩膜技术来忽略电子病历间的数量差异,以精简的自身结构实现不同数量电子病历的统一处理。消融实验证明了时间嵌入方法及编码器稠密网络模型的有效性,同时评估实验表明,经数据预处理、超参数调整后E-TEED-CHD模型的冠心病早期预测准确率为98.71%,千名居民电子病历单轮平均训练时间为17.752 s。E-TEED-CHD模型的准确率及训练速度均优于其他先进模型。
Early detection of coronary heart disease can significantly improve the cure rate,but the early detection rate of coronary heart disease is low due to the inconspicuous prodromal symptoms and lack of self-awareness of patients.By constructing the E-TEED-CHD early prediction model based on the EMR-based time embedding encode-dense network,it is possible to effectively use the non-structured electronic medical record data from residents across regions,institutions,and diseases for early prediction of coronary heart disease.The E-TEED-CHD model embeds the residents'electronic medical records with time position information and parallelly inputs them into multiple encoders for accelerated processing.A dense network model is used to parse the outputs of the encoders and achieve early prediction of coronary heart disease.Furthermore,the E-TEED-CHD model uses a novel medical record masking technique to ignore the quantity differences of electronic medical records among patients,achieving unified processing of different numbers of EMRs with a streamlined structure.Ablation studies demonstrate the effectiveness of the time embedding method and the encode-dense network model.Evaluation experiments show that the early coronary heart disease prediction accuracy of the E-TEED-CHD model,after data preprocessing and hyperparameter tuning,is 98.71%.And the average training time for electronic medical records of 1,000 residents is 17.752 seconds.Both accuracy and training speed of E-TEED-CHD model outperform the other state-of-the-art models.
作者
陈艳章
韩文静
刘增光
CHEN Yanzhang;HAN Wenjing;LIU Zengguang(Univalsoft Joint Stock Co.Ltd.;School of Art,Weifang University of Science and Technology,Shouguang 262700,China;School of Information Engineering,Shandong Vocational College of Science and Technology,Weifang 261053,China)
出处
《软件导刊》
2025年第9期48-54,共7页
Software Guide
基金
潍坊市科学技术发展计划项目(2023GX063)。
关键词
电子病历
时间嵌入
稠密网络
编码器
冠心病预测
electronic medical record
time embedding
dense network
encoder
coronary heart disease prediction
