基于深度学习的“因位性势”四维中医辨证论治算法模型设计

As a precious wealth, the distinguished veteran physicians of TCM (traditional Chinese medicine), for whom the accuracy rate recommended by the diagnosis and prescription on the traditional knowledge base management (KBM) expert system only reaches to 85% approximately, which results a deficiency in information reversion on the clinical syndrome differentiation and treatment and fails to get an excellent clinical application. With the development of the artificial intelligence technology (AIT), AIT is required to be integrated in TCM to exert influence on expanding the range of the disease entities in diagnosis and treatment for the traditional Chinese physicians, improving the ability to diagnose and treat the difficult miscellaneous diseases and shortening the young traditional Chinese physicians’ time accumulation on clinical experience. Based on the two million data on the clinical electronic medical record (EMR) from Sichuan big data platform in TCM, this research integrated our team’s eight-year experience in TCM software development and application to build 48-bit quantization coding of clinical syndrome differentiation and treatment in “etiology, location, nature and tendency of disease” and applied a deep learning of extractive technique in layered features to explore the establishment of artificial intelligence (AI) algorithm and model in the advantages of disease on infantile pneumonia (pneumonic cough) and improve a 5% of accuracy in AI syndrome differentiation and treatment. At the same time, verification was carried out regarding the data on the advantages of disease from 100 distinguished veteran physicians of TCM and the virtual system on the distinguished veteran physicians of TCM to continuously perfect and improve the level of AI in TCM.

名老中医是宝贵财富，但传统知识库管理专家系统诊断和处方推荐的准确率仅能达到85%左右，导致辨证论治信息还原不足，且没有能得到很好的临床应用。随着人工智能技术的发展，中医需要融入人工智能技术，扩展中医生诊疗病种的范围，提升对疑难杂症的诊疗能力，缩短年轻中医生临床经验积累时间等作用。本次研究在四川省中医大数据平台200万份临床电子病历数据基础上，整合团队8年中医软件开发和应用经验，构建辨证论治“因、位、性、势”48位量化编码，采用深度学习分层特征提取技术，探索建立小儿肺炎（肺炎咳嗽）优势病种人工智能算法和模型，提高约5%的人工智能辨证论治的准确率。同时开展100人次名老中医优势病种数据与虚拟名老中医系统的验证，不断完善和提高中医人工智能的水平。

名老中医是宝贵财富，但传统知识库管理专家系统诊断和处方推荐的准确率仅能达到85%左右，导致辨证论治信息还原不足，且没有能得到很好的临床应用。随着人工智能技术的发展，中医需要融入人工智能技术，扩展中医生诊疗病种的范围，提升对疑难杂症的诊疗能力，缩短年轻中医生临床经验积累时间等作用。本项目是建立中医优势病种因位性势深度学习算法模型，完成两类三个算法模型的构建，算法研究已完成，主要有：①因位性势的中医辨证论治算法模型，在中医整体观的指导下本项目对传统辨证方法整理分析，构建基于因、位、性、势的中医辨证论治量化模型，提供支持理法方药一线贯通的中医辨证论治诊断方案。算法模型发表中科院二区论文1篇，影响因子4.098，授权1项发明专利。②提出两种优化深度学习模型应用于中医优势病种（以痛经数据为例），分别为DAELA- LSTM神经网络和Cross-FGCNN神经网络，应用于中医辨证分型，算法准确率分别为95.3%，96.21%，已达到申报书中要求提升正确率5%-10%要求。DAELA-LSTM神经网络算法获得发明专利授权1项，1篇SCI论文投稿中，Cross-FGCNN神经网络算法发表中科院三区论文1篇，影响因子2.96，申请发明专利一项。③以模块的方式已经植入项目组前期开发的中医云健康服务平台，完成算法验证。以成都中医药大学国医馆牵头联盟22家医疗机构，为中医联盟机构提供中医远程会诊平台含中医辨证论治辅助、此算法后台支持、远程会诊及视频服务，联盟机构通过中医远程会诊平台与专家远程视频对接及病历数据共享，让患者在当地机构，足不出市即可享受优质的名中医专家服务，目前已服务243人次。同时为成都中医药大学900余名中医临床跟诊学生及实习生提供了134场远程临床真实案例的中医辨证施治思路及中医药防疫知识学习。