A multi-channel pulse feature optimization method based on traditional Chinese medicine principles

Abstract

The present invention provides a pulse channel optimization method based on the principle of traditional Chinese medicine, comprising the following steps: S1, collecting pulse data of the cun, guan and chi parts of the left and right hands, and performing filtering processing and period division; S2, calculating the importance of the features of each channel, and constructing multiple feature combinations; S3, selecting the feature combination with the highest classification performance at each position, thereby constructing feature subsets of left and right three channels, right hand three channels and both hands six channels; S4, classifying and training the feature subsets. The present invention optimizes the channels of traditional Chinese medicine pulse diagnosis to improve the pulse diagnosis accuracy of various diseases.

Description

Technical Field

[0001] The present invention belongs to the technical field of traditional Chinese medicine diagnosis and machine learning, and specifically relates to a multi-channel pulse feature optimization method based on traditional Chinese medicine principles. Background Art

[0002] Pulse diagnosis in traditional Chinese medicine is one of the most representative diagnostic methods of "looking, smelling, asking, and palpating" in traditional Chinese medicine. Traditional Chinese pulse diagnosis has been widely used in clinical practice. Accurate use of the pulse of the left and right hands is particularly important for judging the physiological and pathological information of patients. Practitioners of traditional Chinese medicine usually use the "three parts and nine signs" method to diagnose the status and disease information of different organs of the human body. Among them, excellent doctors can judge the patient's living habits, physical constitution, previous medical history, etc. through the pulse information they have mastered. However, the different subject and object of this diagnostic method has a great influence on the diagnostic results. The diagnosis of pulse has no quantitative data and lacks objectivity. With the development of sensors and artificial intelligence technology, computer-assisted medicine has been widely used.

[0003] In the existing pulse diagnosis-assisted diagnosis, there are two steps for feature optimization: one is to directly extract the proposed features for research and analysis without performing feature optimization; the other is to directly use the feature selection process in other fields for selection, without combining the theory of the pulse at different positions in traditional Chinese medicine with the conditions of different organs of the human body. Traditional Chinese medicine believes that the left hand inch position is associated with the heart and pericardium, the left hand guan position is associated with the liver and gallbladder, the left hand chi position is associated with the kidneys and bladder, the right hand inch position is associated with the lungs and chest, the right hand guan position is associated with the spleen and stomach, and the right hand chi position is associated with the kidneys and large intestine.

[0004] Traditional Chinese medicine is a mature medical system. The objective pulse diagnosis of traditional Chinese medicine should also be analyzed based on the theoretical methods of the traditional Chinese medicine system. Only in this way can the development efficiency of the objective pulse diagnosis system and the diagnostic performance of related diseases be improved. Therefore, it is necessary to combine traditional Chinese medicine pulse diagnosis and machine learning algorithms to provide an objective, accurate and fast diagnosis method. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a multi-channel pulse feature optimization method based on the principles of traditional Chinese medicine in view of the deficiencies of the above-mentioned prior art. The method is scientifically and reasonably designed, highly practical, simple to operate, can quickly and accurately determine the patient's disease type, and can be promoted for use.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a multi-channel pulse feature optimization method based on the principle of traditional Chinese medicine, characterized in that it includes the following operating steps:

[0007] S1. Collect six-channel pulse signals from the Cun, Guan and Chi parts of the left and right hands of the sample personnel, and pre-process the collected six-channel pulse signal data. First, the collected pulse signal is subjected to noise removal and baseline drift removal by using a Butterworth filter to obtain a high-quality pulse signal. Then, a cycle segmentation algorithm is used to find the starting point of a single-cycle pulse signal. According to the starting point, the single-cycle pulse is identified and segmented. The sample personnel include healthy people and patients with specific diseases.

[0008] S2, respectively extracting pulse features of six channels of pulse signal data after preprocessing of the sample personnel;

[0009] S3, preliminary selection of data features: Use the nearest component analysis method to calculate the importance of the pulse features of the six channels of the sample personnel extracted in S2. According to the importance ranking results, the pulse features of the six channels are preliminarily selected, and the pulse features with high importance in each channel are selected as the basic features of the channel;

[0010] S4, feature optimization: based on the basic features of the same channel selected in S3, according to the number of basic features required to construct the feature combination, construct multiple different feature combinations, send them to different machine learning algorithms, comprehensively select the feature combination and machine learning algorithm with good classification effect, and use the feature combination with good classification effect as the significant pulse feature of the channel;

[0011] S5, model training: based on the different degrees of association between the specific disease and the organs corresponding to the TCM pulse diagnosis position, select several channels of significant pulse features for combination, construct multiple different feature subsets, send the feature subsets to the machine learning algorithm with good classification effect selected in S4 for classification model training, and test the classification effect of the trained classification model to select the feature subset with good comprehensive classification effect, and the channels corresponding to the feature subset with good classification effect are used as the feature subset for disease diagnosis of patients with the specific disease.

[0012] In traditional Chinese medicine, hypertension is divided into four types, namely, liver yang hyperactivity type, yin deficiency and yang hyperactivity type, phlegm dampness obstruction type, and qi stagnation and blood stasis type. Most people's hypertension in life is liver yang hyperactivity type and qi stagnation and blood stasis type. After suffering from hypertension, it may damage the heart, kidneys and cerebral blood vessels of a person, and has a very high incidence rate. Therefore, when studying the pulse of hypertensive patients, the pulse at the inch, guan and chi position of the left hand contains the information of the heart, liver and kidney meridians, which should be more different from that of normal people, and the information of the lung, spleen and life contained in the pulse at the inch, guan and chi position of the right hand should be less different from that of normal people. When optimizing the features of the inch, guan and chi pulses of the left and right hands, considering the above content will improve the classification accuracy and reduce the complexity of data processing. When the patient with the specific disease is a hypertensive patient, based on the close relationship between hypertension and the heart, the pulse at the inch position of the left hand is focused on, and a feature subset corresponding to the three channels of the left hand, the three channels of the right hand and the six channels of both hands is constructed. After the model is trained, it is considered that the feature subset adapted to the hypertension disease is the feature subset corresponding to the three channels of the left hand.

[0013] Compared with the prior art, the present invention has the following advantages:

[0014] 1. The present invention is scientifically and reasonably designed, highly practical, and can effectively improve the speed and accuracy of TCM diagnosis, the development efficiency of the objective pulse diagnosis system, and the diagnostic performance of related diseases.

[0015] 2. The present invention is simple to operate, and rationally applies machine learning and model training in the field of traditional Chinese medicine diagnosis, quickly collects original pulse diagnosis data, and quickly draws diagnostic conclusions, thereby reducing the workload and difficulty of traditional Chinese medicine doctors and improving accuracy.

[0016] The present invention is further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a block diagram of the working principle of the present invention.

[0018] Figure 2 It is a schematic diagram of the operation steps of the present invention.

[0019] Figure 3 It is a schematic diagram of selecting the importance of basic features by the neighbor component analysis in the present invention.

[0020] Figure 4 It is a schematic diagram of sending the feature combination into the machine learning algorithm for evaluation and ranking in the present invention.

[0021] Figure 5 It is a schematic diagram of organs corresponding to different pulse positions of the left and right hands in the present invention. DETAILED DESCRIPTION

[0022] like Figure 1 and Figure 2As shown, the multi-channel pulse feature optimization method based on the principles of traditional Chinese medicine in the embodiment of the present invention is applied to the diagnosis of hypertension, and the specific operation steps are as follows:

[0023] S1. Use a multi-channel piezoelectric sensor pulse acquisition device to collect six-channel pulse signals from the left and right hands of the sample personnel, including healthy people and patients with specific diseases, and filter the collected six-channel pulse signal data based on the Butterworth filter to eliminate noise and remove baseline drift to obtain high-quality pulse signals. Then, use the cycle segmentation algorithm to determine the starting point of the single-cycle pulse, and identify and extract the single-cycle pulse based on the starting point.

[0024] like Figure 5 As shown, according to the principles of traditional Chinese medicine, the Cun, Guan and Chi parts of the left hand correspond to the heart, liver and kidney respectively, and the Cun, Guan and Chi parts of the right hand correspond to the lungs, spleen and life respectively. Patients with specific diseases are patients with hypertension, and Chinese medicine pulse diagnosis for patients with hypertension should focus on these six parts.

[0025] S2, respectively extracting pulse features of six channels of pulse signal data after preprocessing of the sample personnel;

[0026] S3. Preliminary selection of data features: Figure 3 As shown, the importance of the pulse features of the six channels of the sample personnel extracted by S2 is calculated using the nearest component analysis method, and the top six features are extracted according to the importance ranking results. Among them, the important features of the left hand Cun position are the mean, main wave amplitude, dicrotic wave amplitude, main wave time, dicrotic wave time, and main wave to dicrotic wave time; the important features of the left hand Guan position are variance, main wave amplitude, dicrotic wave amplitude, pre-dicrotic wave time, dicrotic wave time, and main wave to pre-dicrotic wave time; the important features of the left hand Chi position are main wave amplitude, dicrotic wave amplitude, main wave time, dicrotic wave time, main wave to dicrotic wave time, and pre-dicrotic wave to dicrotic wave time; the important features of the right hand Cun position are wavelet packet energy characteristics, variance, dicrotic wave amplitude, pre-dicrotic wave time, dicrotic wave time, and main wave to pre-dicrotic wave time; the important features of the right hand Guan position are variance, main wave amplitude, dicrotic wave amplitude, wavelet packet energy characteristics, dicrotic wave time, and main wave to pre-dicrotic wave time; the important features of the right hand Chi position are variance, main wave amplitude, dicrotic wave amplitude, main wave time, dicrotic wave time, and main wave to dicrotic wave time. The pulse features of the six channels are preliminarily selected and combined, and the pulse features with high importance in each channel are selected as the basic features of the channel;

[0027] S4, feature optimization: Figure 4As shown, based on the basic features of the same channel selected in S3, according to the number of 3 basic features required for constructing each channel feature combination set by hypertension disease, multiple different feature combinations are constructed and sent to different machine learning algorithms. 20% of the data are randomly extracted from the data set, and the data are trained and tested using a five-fold cross validation method. According to the classification effect, the corresponding feature combination is analyzed and selected. In this embodiment, three machine learning algorithms, KNN, SVM, and DT, are preferably used. The feature combination and machine learning algorithm with good classification effect for healthy people / hypertensive patients are comprehensively selected, and the feature combination with good classification effect is used as the significant pulse feature of the channel. When DT is used, the left hand Cun position uses the feature combination of main wave time, dicrotic pre-wave time, and main wave to dicrotic pre-wave time, the left hand Guan position uses the feature combination of variance, dicrotic wave amplitude, and main wave to dicrotic pre-wave time, and the left hand Chi position uses the feature combination of main wave time, dicrotic wave time, and main wave to dicrotic wave time. When using SVM, the left hand Cun position uses the feature combination of main wave amplitude, dicrotic wave amplitude, and time from main wave to dicrotic wave, the left hand Guan position uses the feature combination of variance, main wave amplitude, and dicrotic wave amplitude, and the left hand Chi position uses the feature combination of variance, main wave amplitude, and dicrotic wave amplitude. When using KNN, the left hand Cun position uses the feature combination of main wave amplitude, dicrotic wave amplitude, and time from main wave to dicrotic wave, the left hand Guan position uses the feature combination of variance, dicrotic wave amplitude, and dicrotic wave time, and the left hand Chi position uses the feature combination of variance, main wave amplitude, and time from main wave to dicrotic wave. Based on the above results, a feature subset of the three channels of the left hand was constructed, and the healthy person\hypertensive patient data set was used to train and test the model using five-fold cross validation. The performance is as follows:

[0028] As shown in Table 1, it is found that the KNN algorithm has better classification effect on healthy people\hypertensive patients;

[0029]

[0030] Table 1 Comparison of model performance of feature subset training of different algorithms

[0031] S5. Model training: Based on the different degrees of association between the specific disease and the organs corresponding to the TCM pulse diagnosis position, select significant pulse features of several channels for combination, construct multiple different feature subsets, and send the feature subsets to the machine learning algorithm with good classification effect obtained by optimization in S4. The data set constructed by 35 healthy people and 11 hypertensive patients is used to train and test the model by a five-fold cross-validation method. According to the results, the feature subset with good comprehensive classification effect of healthy people\hypertensive patients is selected, and the channels corresponding to the feature subsets with good classification effect are used as the feature subsets for disease diagnosis of patients with the specific disease.

[0032] In S5, a classification model constructed by features in six channels of both hands, three channels of the left hand, and three channels of the right hand is obtained. The three classification models are compared and optimized, as shown in Table 2:

[0033]

[0034] Table 2 Comparison of model performance for feature subset training corresponding to three channels

[0035] For hypertension, according to the evaluation results, the classification model of the feature subset constructed with six-channel features and the classification model of the left-hand three-channel are better than the right-hand three-channel. The number of channels selected for the left-hand three-channel is smaller, and the trained model also has a good classification effect, which is suitable for actual diagnosis scenarios. In other words, it is concluded that the feature subset adapted to hypertension is the feature subset corresponding to the left-hand three-channel.

[0036] The above is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent change made to the above embodiment according to the technical essence of the invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A multi-channel pulse feature optimization method based on the principles of traditional Chinese medicine, characterized in that: The steps include: S1. Collect six-channel pulse signals from the Cun, Guan and Chi parts of the left and right hands of the sample personnel, and pre-process the collected six-channel pulse signal data. First, the collected pulse signal is subjected to noise removal and baseline drift removal by using a Butterworth filter to obtain a high-quality pulse signal. Then, a cycle segmentation algorithm is used to find the starting point of a single-cycle pulse signal. According to the starting point, the single-cycle pulse is identified and segmented. The sample personnel include healthy people and patients with specific diseases. S2, respectively extracting pulse features of six channels of pulse signal data after preprocessing of the sample personnel; S3, preliminary selection of data features: The pulse features of the six channels of the sample personnel extracted in S2 are all calculated for importance, and according to the importance ranking results, the pulse features of the six channels are all preliminarily selected, and the pulse features with high importance in each channel are selected as the basic features of the channel; S4, feature optimization: based on the basic features of the same channel selected in S3, according to the number of basic features required to construct the feature combination, construct multiple different feature combinations, send them to different machine learning algorithms, comprehensively select the feature combination and machine learning algorithm with good classification effect, and use the feature combination with good classification effect as the significant pulse feature of the channel; S5, model training: based on the different degrees of association between the specific disease and the organs corresponding to the TCM pulse diagnosis position, select several channels of significant pulse features for combination, construct multiple different feature subsets, send the feature subsets to the machine learning algorithm with good classification effect selected in S4 for classification model training, and test the classification effect of the trained classification model to select the feature subset with good comprehensive classification effect, and the channels corresponding to the feature subset with good classification effect are used as the feature subset for disease diagnosis of patients with the specific disease.

2. The multi-channel pulse feature optimization method based on the principles of traditional Chinese medicine according to claim 1, characterized in that: In S1, a multi-channel piezoelectric sensor pulse acquisition device is used to collect pulse signals of healthy people and patients with specific diseases.

3. The multi-channel pulse feature optimization method based on the principles of traditional Chinese medicine according to claim 1, characterized in that: In S3, the nearest component analysis method is used to calculate the importance of the pulse signal features of the six channels.

4. The multi-channel pulse feature optimization method based on the principles of traditional Chinese medicine according to claim 1, characterized in that: When the patient with the specific disease is a patient with hypertension, based on the close relationship between hypertension and the heart, attention is paid to the pulse at the inch position of the left hand, and feature subsets corresponding to the three channels of the left hand, the three channels of the right hand, and the six channels of both hands are constructed. After the model is trained, it is considered that the feature subset adapted to hypertension is the feature subset corresponding to the three channels of the left hand.

Images