应用深度学习实现儿科临床疾病智能辅助诊断

Realize the Intelligent Auxiliary Diagnosis of Pediatric Clinical Disease with the Application of Deep Learning

目的:通过对儿科门急诊电子病历的深度学习,研发临床智能辅助诊疗,探讨儿科临床辅助诊疗的可行性。方法:应用循环神经网络(RNN)的深度学习方法对儿科门诊电子病历进行模型训练,通过建立儿科的主数据,推导病人可能的诊断。建立深度学习模型,在儿科专科门诊60万份电子病历及8万份住院病历基础上,根据病历质量筛选出儿科门诊常见的242种疾病,进行模型训练。结果:top-1准确率达到了76.436%(即模型判断中最有可能的诊断和医生的诊断符合),top-3准确率达到92.388%(即模型判断中最有可能的前三个诊断中有一个和医生的诊断符合),top-5准确率达到了95.261%。结论:应用规范化的电子病历实现儿科人工智能辅助诊疗是可行的,并可通过辅助决策分析帮助基层临床儿科医生规范临床诊疗行为。

Objective： Implement Clinical Decision Support System （CDSS） by using deep learning model with outpatient clinical records in pediatrics to prove the feasibility. Methods： Train a Recurrent Neural Network （RNN） model to tagging medical entities; Setup tree-based master data to standardize the information extracted; Train a deep learning model with ultra-high dimension vector as input to make diagnosis, which can cover 242 common diseases. Results： Accuracy is used as evaluation metric, and top1 accuracy is 76.436%, top3 accuracy is 92.388%, top5 accuracy is 95.261%. Conclusion： It is feasible to use deep learning model in CDSS with standardized medical events, and this solution can be introduced into community hospitals where pediatricians are in great deficiency.