Research On Hybrid Intelligent Based Syndrome Differentiation System For Traditional Chinese Medicine

In a long course of struggling against diseases, traditional Chinese medicine(TCM) has been evolved into a unique and integrated theoretical system, and has beenremarkably contributing to the health of people in China and all over the world. TCMhas been clinically observed to have dramatic performance in treating many chronicand systematic diseases such as the treatment of chronic hepatitis B (CHB). So TCMis getting more and more popularity, and attracted the attention of many researcher-s. However, it also has been badly hindered from being popularized and further de-veloped due to the empirical, unquantifiable and obscure features of its diagnostics.Syndrome differentiation in TCM is an important part in the theory of TCM. It is alsoan important basis for making up a prescription and treatment in clinical. The core ofTCM syndrome differentiation is the study of syndrome classification and diagnosticcriteria. However, the current process of TCM syndrome differentiation is lack of astrict designed and unified framework, and a standardized and quantitative diagnosticcriterion. Therefore, it is the issue to study in this article on how to standardize, objec-tify and be computable with computability the syndrome differentiation of traditionalChinese medicine science, which is full of empirical and obscurity.This paper aims to use intelligent techniques to conduct a comprehensive studyabout syndrome from the two perspectives of Chinese medicine and western medicine.In order to design a standardization and objective framework for TCM syndrome d-ifferentiation process, this paper introduces the theory of hybrid intelligent systemto establish a hybrid intelligent based syndrome differentiation for CHB. As current-ly much more research methods applied to TCM syndrome classification, but stillhave not a universally applicable method. On the other hand, due to the complexi-ty and multi-mode of syndromes, the process of syndrome differentiation can not be accurately simulated by one technology. Thereby, it is possible to use theories andmethods of the complexity of scientific research to study syndrome. On the basis ofunderstanding and analyzing the current syndrome differentiation research state andrelating intelligent algorithms, the main achievement of this paper is as follows:1. Multi-view based hybrid feature selectionFeature selection is an important technique of data pre-processing, which is aimedto recognize so as to eliminate the features, in all features, which are redundant or irrel-evant to the issue studied. The dataset of TCM has objective and subjective features,the number is huge. At the same time, collecting data is never an easy job in CHBapplications because of time consuming and costly work. So feature selection is thekey step in the TCM syndrome differentiation. There are so many feature selectionmethods currently. But they can not obtain a comprehensive result of the key fea-tures of syndromes by itself respectively. So in this paper, we propose a Multi-Viewbased Hybrid Feature Selection (MVHFS) method. The proposed method firstly par-tition features of data into different disjoint views according to the nature of features,such as TCM symptoms, TCM signs, and western indicators. And then, the proposedmethod applies hybrid feature selection algorithm, which combines many filters basedfeature selection methods, such as Relief, LVF, mRMR and FCBF, to pick up the keyfeatures of each syndrome on each feature view. The obtained key features of eachsyndrome by proposed method are different each other, which reflects the differencebetween the syndromes, and to lay the foundation for subsequent models of syndromedifferentiation.2. Calculate feature weights combined with distribution informationFeature weight is a subjective evaluation and objectively reflection of compre-hensive measure about the degree of importance of feature. In the field of TCM,the importance and role of the different features to syndrome diagnosis are differen-t. The more important the role of a feature is, the greater its weight should be. Theresearchers often calculate the feature weights according to the occurrence frequen-cy of the features in TCM. They do not consider the distribution information of thefeatures between classes. In this paper, we propose a modified TF-IDF method tocompute the feature weights. We consider the distribution information of the features between classes. Thereby, it can intuitively distinguish the role of different featuresto syndromes. It also quantified shows that the role of the same feature to differentsyndromes is different. This method is consistent with the theory of TCM, and alsolays the foundation for subsequent models of syndrome differentiation.3. Hybrid intelligent syndrome differentiation model based on feature selectionThe essence of TCM syndrome differentiation is syndrome classification. Thereare many classifying methods currently. However, we can not use single classifier orsingle model to improve the classification accuracy of syndrome differentiation due tothe complicated relationships between syndromes and features of TCM and westernmedicine. In additions, it is important to obtain the class probability estimation abouteach patient in the field of TCM diagnosis. accordingly, the doctors can accuratelymake up the prescription and treatment programs for each patient. In this paper, weintroduce the theory of hybrid intelligent system, weighted fuse BayesNet, WPET andWCBA methods, and construct a hybrid intelligent syndrome differentiation modelbased on feature selection. From the experimental results, we can see that this methodcan obtain optimal performance on UCI datasets and CHB dataset. It is shown that thevalidity of the method. We use the proposed method to predict the class probabilityof new180cases, and obtain consistent results with clinical. Further, the proposedmethod shows the potential applications in clinical practice.4. Development of the syndrome differentiation systemWe integrate some of the proposed algorithms to develop a hybrid intelligentbased syndrome differentiation system, which is applied to CHB dataset. By thissystem, we can obtain the optimal feature subsets and predict the syndrome and classprobability of an input case. In the future, this system can fuse new technology forfurther improvement.