肝癌病证计算智能量化研究及体系构建

liver cancer medicine dialectical thinking mode is presented which the overall concept of traditional Chinese medicine as a guide, disease location of internal organs as the core system, disease classification according to the new legislation permits.in this study. This mode of thinking to abandon the drawbacks . During the course, not only syndrome differentiating from signs and symptoms but also combined disease with syndrome are emphasizes. it emphasis on liver cancer disease location, disease characteristics of acquisition and recognized as research-based, always combined with disease resistance to disease location connotation throughout dialectical, the organic combination of the latter two can cover a variety of complex clinical syndromes of liver cancer. It is hard to establish a standardized, accurate mathematical model because of the nonlinear, fuzzy characteristics between the features of disease location, disease nature and syndromes. The purpose of this study is to build a diagnostic knowledge database of liver cancer through sorting out the data from the literatures and the clinical cases and analyzing data,to form a knowledge base through collecting the features of disease location, disease nature of the syndromes from the large samples, and coparing the results with the experts’experiences by soft computation, and eventually to form a new intelligent syndrome differentiation mode of liver cancer which combined the fuzzy quality with the accurate quantity through mining the knowledge deeply.Based on the acquisition and identification of the features of disease location, disease nature of liver cancer sydromes, this new intelligent syndrome differentiation system can reveal the basic laws and scientific principles of TCM syndrome differentiation, help improve the accuracy rates of the clinical liver cancer syndromes .It has important meaning to the standardization, intelligence of TCM Syndrome differentiation.

本课题提出以中医整体观为指导、五脏系统病位为核心、按病性分类立证的“中医肝癌辨证思维模式”，该思维模式突出“以象测藏、从症辨证”的原创思维原则，强调以肝癌病位、病性特征的获取和识别为研究基础，以病位结合病性的内涵贯穿辨证始终，两者有机组合后可涵盖临床各种错综复杂的肝癌病证。由于病位、病性特征与病证之间具有非线性、模糊性特征，较难建立规范、准确的数学模型。本项目从肝癌病证入手，通过对搜集的文献及临床病例数据进行整理、分析，从大样本中提取肝癌病证的病位、病性特征，利用计算智能方法建立肝癌病证诊断知识库；对知识库进行挖掘，最终形成模糊定性与精确定量相结合的肝癌病证辨证模型。以肝癌病位、病性特征的获取、识别为研究基础的智能辨证体系研究，在整体上揭示了中医辨证的基本原理与普遍规律，对提高临床诊治水平起到重要的辅助作用，对促进中医辨证的规范化、智能化具有重要的意义。

肝癌病证病因病机复杂，临床症状繁杂多变，辨证分型较多且不统一，各医家对肝癌病位、病性采用的诊断标准和辨证方法不尽一致，这就给肝癌辨证诊断的标准化和规范化带来困难。利用人工智能技术建立肝癌中医诊断智能化模型，能够为肝癌中医临床提供数字化和智能化的诊断，以期提高中医肝癌辨证分型的准确性和科学性。肝癌是复杂的全身疾病，临床诊断辨证具有多病位、多病性的特点，采用传统的神经网络、决策树等方法难以较好地处理肝癌证型兼夹的情况。本课题利用多标记学习方法对中医肝癌辨证智能化进行探索，分别采用ML-DT、ML-KNN、ML-RF等多种方法构建肝癌辨证模型，并进行模型之间的比较。从研究结果可以看出：ML-RF模型的Average precision为0.9186±0.0132，其反映了模型诊断结果与实际诊断结果的一致性，这也是评价模型效果的重要指标，在几个模型中ML-RF的Average precision明显优于其他指标；Hamming loss为0.0216±0.0013，其反映了诊断结果中各个子项与真实诊断结果对应情况；One error为0.0148±0.0102，其反映了模型对主证诊断的误判率；Coverage为0.1953±0.0082反映了模型诊断结果与真实诊断结果相比的冗余情况；Ranking loss为0.0016±0.0003，其反映了模型诊断结果中各个子项的排序与真实诊断结果对应情况的误差率。综合上述结果，可以认为ML-RF的综合表现优于其他的算法模型。在今后的研究中，可以将模型进一步扩展推广应用到临床诊疗中，辅助中医师在临床开展肝癌病证的诊疗，实现肝癌临床诊断的客观化和智能化，以期能够提高临床工作效率和临床疗效，并对中医诊断现代化的发展起到一定的促进作用。