基于名老中医经验的中医智能诊疗模型构建研究

Traditional Chinese medicine intelligent treatment is the trend of medical development in the future, the key lies in the diagnosis and treatment of TCM of the Chinese medicine diagnosis and treatment experience, uncertainty, fuzziness is converted into linear logic. With the development of artificial intelligence is becoming more and more mature, a variety of models can simulate the human brain processing information for black box operation and nonlinear adaptive learning. To this end, we put forward the hypothesis: the use of deep reinforcement learning model can be constructed in line with the characteristics of TCM diagnosis and treatment of intelligent diagnosis and treatment system and self learning. In TCM experience as an example, the establishment of kidney disease database; in the tensorflow system environment, using MLP (multilayer perceptron) deep learning method to establish the young model, then use DQN (depth of reinforcement learning network) improved model. Finally, a new model of intelligent diagnosis and treatment of TCM, which is suitable for the diagnosis and treatment of TCM, is developed. The study will help to promote the application of artificial intelligence in the diagnosis and treatment of TCM. It is of great significance in the preservation and excavation of the experience of diagnosis and treatment of traditional Chinese medicine.

中医智能化诊疗是未来医疗发展的趋势，目前中医智化诊疗关键在于中医诊疗的经验性、不确定性、模糊性难以转化为线性逻辑。随着人工智能的发展日渐成熟，多种模型可模拟人脑进行黑箱操作处理非线性信息并可自适应学习。为此，我们提出假说：利用深度强化学习模型可以构建符合中医诊疗特点的智能化诊疗系统并自我学习提高。以中医肾病经验为例，建立肾病病例数据库；在tensorflow系统环境下，先利用MLP（多层感知器）深度学习方法建立雏模型，再利用DQN（深度增强学习网络)完善提高模型。最终研制适应中医诊疗特点的中医智能化诊疗新模型和智能体。该研究将有助于推进人工智能在中医诊疗上的应用，在保存和挖掘名老中医诊疗经验和提高中医诊疗水平上具有重大意义。

本课题围绕中医智能辨证模型开展了探索研究，首先是围绕项目任务书中提出的创立一个肾病的中医智能辨证模型开展了一系列的探索研究，共收集李培旭诊疗的肾病病例1150例，从中标注训练样本3000个，训练出59*500*500*500*18的DNN模型。经测试样本测试精准度达到97%，F1值达到95%，召回率达到95%。其次，由于DNN模型只是一个局域模型，局域模型是无法解决临床的广域问题的，比如多种合并疾病、合并症状问题。为解决这个问题，我们尝试了多种模型的探索，包括RNN和CNN以及知识图谱算法等，均因训练样本、算力、广域扩展性等等问题而没有成功。 最后,经过我们反复探索，自创了一套广域的中医辨证算法，该算法以中医基本理论为基础，采用模糊数学和矩阵分析方法，构建出从症状到证型的分类预测模型，并在计算机上成功实现，以此算法为基础，成功开发出老中医经验保存和共享系统、基于专家经验的专家系统和基于中医教材的中医辅助决策诊疗系统，并应用于中医临床科研一体化系统中。本次研究结果成功解决了中医临床数据即时结构化和数字化问题，解决了中医大数据应用中的中医算法问题，解决了名老中医经验快速推广和应用问题，预期对中医行业发展将产生积极促进作用。