基于深度神经网络学习的余瀛鳌治疗肾病辨证论治辅助决策模型的研究

Dr. Yu Yingao, a well-known doctor of TCM with high prestige in Beijing, has been devoting himself to diagnose and treat nephritis for many years and therefore has developed a syndrome differentiation diagnosis-treatment and corresponding formulas of his own. The key to improve nephritis treatment efficacy is to learn from Dr. Yu as well as to develop our own treatment. A research is done to over 6000 clinical cases of nephritis diagnosed and treated by Dr. Yu. Based on deep learning and data mining, Dr. Yu Yingao’s syndrome differentiation diagnosis-treatment model of nephritis was set up. In this research, the maximum efficacy population is used to mine the development rule and the syndrome feature, i.e. exterior deficiency with interior excess; Automatic time-space、multiple-retrieval. Meanwhile, deep neural network learning is also able to build the contracture, dose relationship and dynamic relation of the effective prescription and the dynamic relation, self-feeling and effective patient feature of the corresponding population. Then, we can obtain the underlying symptom-syndrome-prescription relationship of Dr. Yu Yingao treating nephritis. Thus, a self-organizing map of Dr. Yu’s syndrome differentiation diagnosis-treatment model could be drawn to show the interrelations of each part of the model. In addition, this research was dedicated to reveal the symptoms and development of nephritis, as well as the features of nephritis—exterior deficiency with interior excess; Automatic time-space、multiple-retrieval. Above all, this research’s destination was to uncover the routine of Dr. Yu dignosing and treating patients with nephritis and the formulas he adopts. Learning from Dr. Yu will shed light on the improvement of nephritis treatment efficacy among doctors of TCM.

“首都国医名师”余瀛鳌先生治疗肾病有多年的临床经验积累，存在大量个体治疗显效的患者群，形成了中医药治疗肾病的有效辨证论治方案和有效通治方，是中医临床药治疗肾病的宝贵财富。如何在借鉴余瀛鳌先生治疗肾病的临床经验基础上继承创新，是提高中医药治疗肾病效果的关键。为此，本研究以6000例余瀛鳌先生治疗肾病的临床实际诊疗数据为基础，采用深度神经网络的学习方法，以最大疗效为前提，探索肾病证候演变规律及证候“内实外虚、动态时空、多维界面”的特征，探索有效通治方构成、剂量匹配、动态加减等，探索对应适宜人群动态变化、自我感受、有效患者特征等，发现余瀛鳌先生治疗肾病“症-证-方”潜在相关关系，构建余瀛鳌先生治疗肾病辨证论治的辅助决策模型，建立余瀛鳌先生中医临床实际诊疗路径中辨证论治的自组织映射关系，进而为年轻医生临床辅助开方提供决策支持，也为借鉴余瀛鳌先生临床经验进而提高中医药治疗肾病的诊疗水平提供可行途径。

项目完成了6837份病例录入，数据维度包括患者人口学信息、现病史、既往史、症状、中西医诊断、治疗方案、转归等信息； 通过传统统计学方法，初步统计了余瀛鳌先生临床治疗肾病的用药单因素分析，包括加减用药和处方的频次等；通过深度神经网络的大数据学习方法，实现肾病临床实际诊疗数据的建模分析，同时召开三轮的专家论证会，对模型结果的解读进行了修订，进一步完善模型学习的性能；完成余瀛鳌先生辨证论治辅助决策可视化平台搭建工作，通过输入病症，舌诊，脉诊及治法职责等相关数据，可以得到用药和剂量的建议，准确率达到80%。