基于Co-LSTM-FC网络的糖尿病伴周围神经病变临床决策支持系统研究-《中国医学物理学杂志》

文章信息/Info

Title:: Clinical decision support system for diabetic peripheral neuropathy based on Co-LSTM-FC network

作者:: 刘玉良1; 丁永川1; 郭宇佳1; 赵耕2; 杨伟明1; 1.天津科技大学电子信息与自动化学院，天津 300202； 2.天津医科大学代谢病医院检验科，天津 300070

Author(s):: LIU Yuliang1; DING Yongchuan1; GUO Yujia1; ZHAO Geng2; YANG Weiming1; 1. School of Electronic Information and Automation, Tianjin University of Science and Technology, Tianjin 300202, China 2. Department of Laboratory, Metabolic Disease Hospital of Tianjin Medical University, Tianjin 300070, China

Keywords:: Keywords: diabetic peripheral neuropathy clinical decision support system FC-LSTM network ConvLSTM network

摘要:: 为了实现辅助医生对糖尿病伴周围神经病变（DPN）进行早期诊断与决策，针对DPN早期预测提出一种基于Co-LSTM-FC网络的临床决策支持系统（DPN-CDSS）。Co-LSTM-FC网络模型使用FC-LSTM网络和ConvLSTM网络共同对患者的临床数据进行特征提取，减轻单一模型运算时出现的权重偏差，同时利用全连接神经网络对患病特征进行分类，提高预测模型准确率。本文方法的准确率、特异度、F1值、G-mean值和AUC值分别为95.51%、94.24%、95.06%、95.08%和94.37%，与对比模型相比获得的结果准确度更高。DPN-CDSS用户界面包括用户登录、数据输入和结果显示界面，方便医生和患者进行使用。该系统可提前筛查患者的得病情况，辅助医生进行初期诊断，提升诊疗效率。

Abstract:: Abstract: A clinical decision support system (DPN-CDSS) based on Co-LSTM-FC network is proposed for the early prediction of diabetic peripheral neuropathy (DPN), thereby assisting doctors in the early DPN diagnosis and decision-making. Co-LSTM-FC network model innovatively uses FC-LSTM network and ConvLSTM network to jointly extract the features from the clinical data, which reduces the weight deviation that occurs in the calculation of a single model. Meanwhile, the fully connected neural network is adopted to classify the characteristics of the disease for improving the accuracy of the prediction model. The accuracy, specificity, F1 value, G-mean value and AUC value of the proposed method for DPN prediction are 95.51%, 94.24%, 95.06%, 95.08% and 94.37%, respectively, and the accuracy is higher as compared with other models. Moreover, DPN-CDSS user interface which includes user login, data input and result display interface is convenient for doctors and patients to use. The system can screen for DPN in advance, assist doctors in the initial diagnosis, and improve the efficiency of diagnosis and treatment.