This article is selected from the journal of the Chinese Academy of Engineering "Engineering" No. 1, 2022
Authors: Wang Jie, Duan Lian, Li Hongzheng, Liu Jinlei, Chen Hengwen
来源：Construction of an Artificial Intelligence Traditional Chinese Medicine Diagnosis and Treatment Model Based on Syndrome Elements and Small-Sample Data[J]. Engineering,2022,8(1):29-32.
Editor's note
With the continuous breakthrough of artificial intelligence technology, TCM experts began to mine TCM big data for related research, but unlike modern medicine, which has nearly unified clear diagnosis and treatment rules, there is a crossover between diversified traditional TCM diagnosis and treatment theories, and the diagnosis and treatment rules of different TCM schools and different TCM physicians also cross, resulting in the difficulty of storing high-quality TCM big data with the same rule system. Therefore, in the face of the real situation such as small-sample training or even zero training samples, the use of regular explainable models can be considered to complete the knowledge + artificial intelligence fusion under the combination of disease evidence.
The first issue of Engineering, a journal of the Chinese Academy of Engineering, in 2022, published an article entitled "Using Syndrome Elements to Build a Traditional Chinese Medicine Artificial Intelligence Diagnosis and Treatment Model" by Professor Wang Jie of Guanganmen Hospital of the Chinese Academy of Chinese Medical Sciences and other researchers. This paper takes coronary heart disease as a research breakthrough of integrated traditional Chinese and Western medicine + artificial intelligence for a single disease, introduces artificial prior knowledge, analyzes its possible symptom elements through clinical symptoms and tongue pulse and other information, and then obtains reliable prescription drugs with symptom elements as the core, thereby forming a TCM artificial intelligence diagnosis and treatment model design based on symptom elements, and using small sample diagnosis and treatment data to train this model, which can achieve higher artificial intelligence prescription accuracy than the diagnosis and treatment model formed by "TCM big data".
I. Introduction
Integrative medicine has advantages in the treatment of chronic and difficult diseases. In this new crown epidemic, the utilization rate of traditional Chinese medicine and the total effective rate of treatment of confirmed cases in Hubei Province have exceeded 90%, and the efficacy of integrated traditional Chinese and Western medicine has been affirmed. However, the inheritance of TCM diagnosis and treatment thinking and the future development of integrated TCM still face severe challenges. Since the 50s of the 20th century, people have tried to use information technology to objectiveize the four diagnoses of traditional Chinese medicine; Since the 70s, attempts have been made at home and abroad to develop a TCM expert system with expert clinical diagnosis and treatment thinking, but it has never been able to better simulate the dialectical treatment process of TCM practitioners. Since the 90s, with the continuous breakthrough of artificial intelligence (AI) technology, TCM expert systems have begun to use neural network fuzzy logic, relational databases and other technical methods to mine TCM big data, and carry out objective, standardized and quantitative research. However, learning model parameters based entirely on data-driven methods makes it difficult to integrate the experience and knowledge formed by long-term clinical summaries of TCM practitioners into the model, so it is difficult to impose effective constraints on the model learning process.
In recent years, artificial neural networks that are good at dealing with nonlinear problems, decision trees and random forest algorithms that are good at processing high-dimensional data have all been applied to the construction of TCM models. However, unlike modern medicine, which has nearly unified and clear diagnosis and treatment rules, there are crossovers between diversified traditional TCM diagnosis and treatment theories, and the diagnosis and treatment rules of different TCM schools and TCM physicians also cross, making it difficult to reserve high-quality TCM big data with the same rule system. In the face of the real situation such as small-sample training or even zero training samples, you can consider using a regular explainable model to complete the knowledge + AI fusion under the combination of disease evidence.
Second, the use of symptom elements can unify different diagnosis and treatment rules
Syndrome is the core of the construction and clinical application of the theoretical system of dialectical treatment in traditional Chinese medicine, which is the overall summary of the pathological and physiological changes of human diseases through etiology, location and pathology, and is the result of "differentiation" (summarizing and summarizing the characteristics of symptom groups) and the basis of "treatment" (determining treatment principles and specific treatment methods), which has the characteristics of multi-level, complex, abstract and high-dimensional. However, the scope of use of each dialectical program has problems such as blurred distinctions, high overlap of application scenarios, and cross-dialectical content. This leads to insufficient accuracy of prescriptions based on the TCM expert system, which brings great difficulties to the standardization of diagnosis and intelligent research of TCM.
In view of this, in the artificial intelligence research combining disease symptoms, the symptom elements can be used as the cornerstone to link the internal elements such as symptoms, treatments, traditional Chinese medicines, and prescriptions in the process of diagnosis and treatment. Any complex syndrome is cross-determined by specific dimensions such as disease position, pathology, and evil relationship, and the specific performance of each dimension can be considered as the elements of the dialectical dimension. It is found that about 60 basic dialectical elements of symptoms can be separated by dimensionality reduction and upgrading, and the basic dialectical elements can be arranged and combined, which can cover the types of TCM syndromes. Just as letters are the smallest units that make up a word, the elements of symptoms in the field of TCM are the smallest units that make up symptoms. Different from the complex and high-dimensional syndrome, each symptom element has a specific symptom group, and also has specificity different from other elements, and through combination and superposition, the clinical syndrome can be formed, that is, the symptom element has the characteristics of low dimension, superposition and combination. Each symptom element has its corresponding treatment, traditional Chinese medicine or drug combination, and the combination form of symptom elements and symptom elements is clarified, and the treatment and prescription medicine can be determined accordingly.
Taking coronary heart disease as an example, the diagnosis and treatment process of traditional Chinese medicine with symptom factors as the core was analyzed. All clinical manifestations of patients with coronary heart disease can be attributed to eight major syndrome elements: blood stasis, qi deficiency, phlegm turbidity, yin deficiency, qi stagnation, yang deficiency, cold coagulation, and heat containment, which can become qi deficiency and blood stasis, qi and yin deficiency, sputum and stasis intertwined, qi stasis and blood stasis, sputum resistance heat and yang deficiency and cold coagulation. According to the correspondence between the symptoms and the treatment, it can be seen that the TCM treatment methods of coronary heart disease can include blood circulation, qi invigoration, phlegm dissolving, nourishing yin, rationalizing qi, warming and yang, dispersing cold, clearing heat, etc. According to the treatment method, the corresponding formulas, medicinal pairs or single-flavor Chinese medicines can be obtained, such as coronary No. 2 formulas, codonopsis-astragalus, cubeb-galangal and other medicinal pairs, as well as danshen, Panax notoginseng and other single-flavored Chinese medicines; Then the drug combination is carried out according to the combination form of the syndrome elements, which can correspond to the representative prescription drugs for the treatment of coronary heart disease such as Blood Fu Zhuyu Soup, Gua Xue Bai Banxia Soup, etc. It can be seen that the superposition and combination of symptom elements is the basis and key to the addition and subtraction of drug prescription changes, which can integrate and unify the TCM diagnosis and treatment rules from different sources.
Third, use algorithms to integrate TCM data models suitable for small samples
Algorithm is an important technology in the field of artificial intelligence and an important path for studying artificial intelligence systems in Chinese medicine. In the selection of TCM AI algorithms, we should carefully consider their applicability. For the identification of four-diagnosis syndrome information, considering the number of cases, Bayesian networks and support vector machines can be selected, because they perform well on small-scale data and are more robust to data noise. However, when the algorithm is applied to real-world large data samples, the calculation difficulty increases, which affects the practicality.
The complexity of the TCM syndrome system determines that its research should start from nonlinear design, artificial neural network with its excellent nonlinear problem analysis ability, can simulate the structure and function of the human brain neural network, effectively process the data, identify the hidden complex laws in the data, and have advantages in determining the TCM syndrome problem. However, the neural network algorithm has high requirements for the standardization level of data, and the training needs to have a large scale of data, and the current standardization level of TCM data is insufficient, resulting in the lack of a large amount of training data, which is an important factor restricting the role of this algorithm.
The complete simulation of TCM diagnosis and treatment process and the processing of multi-dimensional and complex data have always been difficult problems in TCM AI research. Both decision trees and random forest algorithms are good at processing high-dimensional data, exploring the interaction between the characteristics of the data, and meeting the requirements of the data. However, decision trees tend to overfit, which can lead to reduced accuracy. The random forest algorithm is to build multiple decision trees in a random way, and is good at dealing with unbalanced data. The research on the establishment of prediction model for chronic fatigue TCM syndrome elements using random forest algorithm has achieved high prediction accuracy. However, its controllability is insufficient, and the model cannot classify small-scale data well, and there is also a problem of limited data sources in the field of traditional Chinese medicine.
Although artificial intelligence has achieved some valuable results in the field of traditional Chinese medicine, there are two main problems. First, there is the lack of standardized and objectified data. AI computing requires a large amount of standardized and objectified data, but limited by the particularity of the TCM industry, the collected data flow is significant in terms of differentiation and subjectivization, which makes the objectification process of TCM slow, resulting in the lack of a large amount of data support for the above-mentioned artificial neural networks and other new artificial intelligence algorithms, and it is difficult to make breakthroughs. Second, the application scenario of existing research is single, and although only trying to apply one algorithm can realize the construction of a predictive model based on certain characteristics of data, the versatility and portability are generally insufficient to form a systematic research result that is common to multiple diseases and realizes the complete diagnosis and treatment process of symptom-symptom-treatment-prescription medicine. Therefore, according to the current situation of TCM AI research, on the basis of solving the problem of data standardization, a variety of algorithms are combined to achieve accurate calculation from symptoms to symptoms, so as to realize the whole diagnosis and treatment process of symptom-symptom-treatment-prescription medicine, which is an urgent problem to be solved.
Fourth, the research strategy and direction of artificial intelligence in integrated traditional Chinese and western medicine
On the basis of the new generation of artificial intelligence algorithm, we propose the idea of exploring a regularized TCM AI research model based on the symptom elements: taking TCM AI as the research object, taking the symptom elements as the entry point, taking the rule integration as the path, and taking the symptom points given by the industry diagnosis and treatment guidelines as the weight, through the process of symptom combination, disease identification, symptom derivation, rule of law, prescription generation, addition and subtraction of medicinal flavor, and drug quantity assignment to form a recommended prescription (Figure 1). In addition, a small-sample, high-quality TCM diagnosis and treatment data was used to train the model to correct the accuracy of symptom calculation. Feedback mechanisms can also be leveraged to allow TCM physicians to make adjustments based on the output results and add such adjustments to the rule dataset as new rules.
Secondly, with the help of the learnability and scalability of the artificial intelligence model of traditional Chinese medicine, the contents of symptoms, syndrome elements, traditional Chinese medicines, prescriptions, clinical cases, classical medical books and medical cases are considered, so that the knowledge system of TCM diagnosis and treatment from different sources can complete the integration of knowledge system, establish the corresponding relationship between symptom-symptom elements-symptom-treatment-prescription-Chinese medicine, and improve the accuracy of dialectical pointing. In terms of specific implementation, it can be considered to use symptom and symptom elements as nodes, and symptom-symptom element correlation as the boundary to construct a knowledge graph of diagnosis and treatment rules. Using the symptom integral given by the single disease industry diagnosis and treatment guidelines as the weight, the Laplacian matrix is calculated by constructing the adjacency matrix and degree matrix to represent the weight of different symptoms under the symptom elements, and then using the convolution operation to weighted the sum output results, so as to establish a visual model, on the basis of which the visual display of rules is realized.
Based on the above design ideas, it is committed to forming a TCM artificial intelligence system that conforms to industry standards, classic ancient books and expert opinions (Figure 1), which provides a new direction for the research of standardized and objectified TCM artificial intelligence diagnosis and treatment system.
Figure 1. Design route of artificial intelligence model integrating traditional Chinese and Western medicine for single disease. (a) Acquisition and standardization of TCM terminology. By obtaining the terminology information of symptoms and signs from different sources, the patient's tongue, pulse, face, visual image, voice and other information are collected, and the above information content is labeled with data, and then the data standardization work is completed. (b) Acquisition of TCM rules of diagnosis and treatment. TCM diagnosis and treatment rules can come from TCM clinical diagnosis and treatment guidelines, TCM expert experience, TCM textbooks, ancient TCM textbooks, etc., and the diagnosis and treatment rules from different sources above are associated and integrated through symptom elements, so as to form integrated diagnosis and treatment rules with symptom elements as the core. (c) Taking the construction of graph convolutional neural network and knowledge graph as an example, the scheme of integrating algorithm rules and forming visual models is illustrate. The TCM artificial intelligence knowledge graph is constructed, taking the symptom elements and symptoms as nodes, the correlation between symptoms and symptoms as the boundary, and the symptom points given by the disease industry diagnosis and treatment guidelines as the basis to represent the weight of different symptoms under the symptom elements, and then use convolution operation to weighted sum the output results, and support visual display. (d) Integrate the rules of diagnosis and treatment from various sources to form a process of identifying symptoms and exporting prescriptions. Taking standardized symptoms as the input layer, the judgment of standardized symptom corresponding symptom elements is first completed, and then the syndrome is formed through the superposition and combination of symptom elements, and the prescriptions and traditional Chinese medicines corresponding to the combination of main symptom elements are output through the correspondence relationship of symptom-treatment. In addition to the main symptom elements, other symptom elements and symptoms are output with corresponding traditional Chinese medicines, forming a drug addition and subtraction prescription for the main prescription as the output layer. (e) Realize the visual display of rules based on knowledge graph and prescription recommendation results.
V. Outlook
We take coronary heart disease as the research breakthrough of integrated traditional Chinese and Western medicine + AI for a single disease, introduce artificial prior knowledge, analyze its possible symptom elements through clinical symptoms and tongue pulse and other information, and then obtain reliable prescription drugs with symptom elements as the core, thus forming a TCM artificial intelligence diagnosis and treatment model design based on symptom elements, and using small sample diagnosis and treatment data to train this model, which can achieve higher AI prescription accuracy than the diagnosis and treatment model formed by "TCM big data".
We expect that the TCM diagnosis and treatment system will realize the development from a single-disease diagnosis and treatment model to a multi-disease diagnosis and treatment model, and from a pure TCM diagnosis and treatment system to a true integrated TCM diagnosis and treatment system. We hope that in the near future, it can simplify the consultation process, improve the rate of diagnosis and treatment, and allow patients to easily receive artificial intelligence diagnosis and treatment services based on the scarce high-level diagnosis and treatment experience of high-level TCM experts. Even if you are far from the hospital, you can receive the best quality authentic medicinal materials at home and receive the best quality medical services.
Note: The content of this article has been slightly adjusted, if necessary, you can view the original text.
Adapted from the original text:
Jie Wang, Lian Duan, Hongzheng Li, Jinlei Liu, Hengwen Chen.Construction of an Artificial Intelligence Traditional Chinese Medicine Diagnosis and Treatment Model Based on Syndrome Elements and Small-Sample Data[J]. Engineering,2022,8(1):29-32.
Note: The paper reflects the progress of research results and does not represent the views of China Engineering Science magazine.