Traditional Chinese medicine model training method and traditional Chinese medicine diagnosis equipment

Abstract

The invention relates to a traditional Chinese medicine model training method and traditional Chinese medicine diagnosis device.The traditional Chinese medicine model training method comprises the steps that clinical data are obtained, and the clinical data comprise electrocardiosignals, fingertip photoelectric pulse wave signals and wrist pulse waveforms; obtaining a diagnosis result corresponding to the clinical data; taking the diagnosis result as a label of the clinical data, and generating training data according to the clinical data and the label; and training a traditional Chinese medicine model by using the training data to obtain a pre-trained traditional Chinese medicine model. The traditional Chinese medicine model is generated.

Description

Technical Field

[0001] The present invention relates to the technical field of traditional Chinese medicine diagnosis equipment, and particularly relates to a method for training a traditional Chinese medicine model and a traditional Chinese medicine diagnosis equipment. Background Art

[0002] In traditional Chinese medicine theory, deficiency, cold, dampness, congelation, and stasis respectively represent different stages of the disease course. Traditional Chinese medicine diagnosis mainly relies on traditional Chinese medicine doctors' methods of feeling the pulse, face diagnosis, tongue diagnosis, and inquiry of patients, which is highly subjective, resulting in poor consistency and unable to meet the requirements of modern evidence-based medicine. Eventually, it leads to a lack of standards for treatment plans and uneven treatment effects. Summary of the Invention

[0003] In view of the problem of how to improve the consistency of traditional Chinese medicine diagnosis, the present invention provides a method for training a traditional Chinese medicine model and a traditional Chinese medicine diagnosis equipment. The traditional Chinese medicine model trained by the method for training a traditional Chinese medicine model can provide digital indicators to assist in improving the consistency of traditional Chinese medicine diagnosis.

[0004] In a first aspect, the present application provides a method for training a traditional Chinese medicine model, including:

[0005] Obtaining clinical data, where the clinical data includes: electrocardiogram signal, fingertip photoplethysmogram signal, and wrist pulse waveform;

[0006] Obtaining a diagnosis result corresponding to the clinical data;

[0007] Using the diagnosis result as a label for the clinical data, and generating training data according to the clinical data and the label;

[0008] Training a traditional Chinese medicine model using the training data to obtain a pre-trained traditional Chinese medicine model.

[0009] In some embodiments, the diagnosis result consists of at least one of deficiency, cold, dampness, congelation, and stasis.

[0010] In some embodiments, the generating training data according to the clinical data and the label includes:

[0011] Performing noise reduction and band-pass filtering on the electrocardiogram signal;

[0012] Calculating the number of R waves of the electrocardiogram signal within a preset duration to obtain a heart rate parameter; wherein, the preset duration is greater than or equal to 30 seconds and less than or equal to 2 minutes;

[0013] Using the heart rate parameter and a first diagnosis result as training data, where the first diagnosis result includes: deficiency and non-deficiency.

[0014] In some embodiments, generating training data based on the clinical data and labels includes: calculating the variation law of the R-wave amplitude of the electrocardiogram signal to obtain respiratory parameters;

[0015] Using the heart rate parameter and the first diagnosis result as training data includes: using the heart rate parameter, the respiratory parameter, and the first diagnosis result as training data.

[0016] In some embodiments, generating training data based on the clinical data and labels includes:

[0017] Obtaining the front and back N effective pulse wave amplitudes adjacent to the maximum amplitude waveform of the wrist pulse waveform; where N is greater than or equal to 8 and less than or equal to 20;

[0018] Using the N effective pulse wave amplitudes and the second diagnosis result as training data, where the second diagnosis result includes: cold and non-cold.

[0019] In some embodiments, generating training data based on the clinical data and labels includes:

[0020] Obtaining TP, LF, HF, SDNN, SDANN, and RMSSD of heart rate variability from the fingertip photoelectric pulse wave signal;

[0021] Using the TP, LF, HF, SDNN, SDANN, RMSSD, and the third diagnosis result as training data, where the third diagnosis result includes: damp and non-damp.

[0022] In some embodiments, generating training data based on the clinical data and labels includes:

[0023] Obtaining the time-frequency diagram corresponding to the fingertip photoelectric pulse wave signal;

[0024] Feeding the time-frequency diagram into a pre-trained convolutional neural network, and the pre-trained convolutional neural network outputs the peripheral microcirculation blockage parameter;

[0025] Using the peripheral microcirculation blockage parameter and the fourth diagnosis result as training data, where the fourth diagnosis result includes: coagulation and non-coagulation.

[0026] In some embodiments, generating training data based on the clinical data and labels includes:

[0027] Obtaining the time difference between the R-wave peak of the electrocardiogram signal and the peak of the fingertip photoelectric pulse wave signal;

[0028] Calculating the blood flow propagation speed parameter according to the time difference, the propagation speed of the electrocardiogram signal, and the propagation speed of the fingertip photoelectric pulse wave signal;

[0029] Use the blood flow propagation speed parameter and the fifth diagnostic result as training data, where the fifth diagnostic result includes: stasis and non-stasis.

[0030] In a second aspect, the present application provides a traditional Chinese medicine diagnostic device, which includes:

[0031] A clinical data acquisition device to be diagnosed, which is used to acquire clinical data to be diagnosed;

[0032] A memory that stores a traditional Chinese medicine model; the traditional Chinese medicine model is trained by using the traditional Chinese medicine model training method described above.

[0033] A processor, connected to the memory, and the processor is configured to call the traditional Chinese medicine model to output a diagnostic result according to the clinical data to be diagnosed.

[0034] In an embodiment, the traditional Chinese medicine diagnostic device further includes:

[0035] An interaction device, connected to the processor, and the interaction device is used to guide the user to acquire the clinical data to be diagnosed by the user; and display the diagnostic result.

[0036] The present invention is the first to combine and match physiological parameters with traditional Chinese medicine pathology, realizing the quantification and digitization of traditional Chinese medicine; by using electrocardiogram signals, fingertip photoplethysmogram, wrist pulse waveforms, and the diagnostic results of traditional Chinese medicine doctors, training data is generated, and then a traditional Chinese medicine model is trained to find the potential relationships between the key parameters of physiological detection data and different stages of the development of traditional Chinese medicine courses, including deficiency, cold, dampness, stasis, and coagulation, enabling objective quantification and digitization of traditional Chinese medicine diagnosis and improving the consistency of traditional Chinese medicine diagnosis. Description of the Drawings

[0037] Figure 1 It is a flowchart of an embodiment of the traditional Chinese medicine model training method of the present application;

[0038] Figure 2 It is a waveform schematic diagram of an electrocardiogram signal;

[0039] Figure 3 It is a waveform schematic diagram of a fingertip photoplethysmogram signal;

[0040] Figure 4 It is a schematic diagram of a wrist pulse waveform;

[0041] Figure 5 It is a flowchart of an embodiment of the traditional Chinese medicine model application method of the present application. Detailed Embodiments

[0042] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present invention.

[0043] In the description of the present invention, the terms "first" and "second" are only used for descriptive purposes, and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present invention, "a plurality" means two or more, unless otherwise specifically defined.

[0044] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration, or explanation". Any embodiment described as "for example" in the present invention is not necessarily construed as being more preferred or having more advantages than other embodiments. In order for any person skilled in the art to implement and use the present invention, the following description is given. In the following description, details are set forth for the purpose of explanation. It should be understood that those of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other instances, well-known structures and processes are not elaborated in detail to avoid unnecessary details from obscuring the description of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope that conforms to the principles and features disclosed in the present invention.

[0045] Example 1

[0046] Considering that there is no direct theoretical connection between traditional Chinese medicine theory and modern medical detection indicators, the AI machine learning method of this application can efficiently and accurately find the direct relationship between modern medical physiological indicators and traditional Chinese medicine diagnosis compared with the logic analysis algorithm designed manually.

[0047] In a first aspect, this application provides a method for training a traditional Chinese medicine model. The method for training a traditional Chinese medicine model extracts characteristic parameters respectively using physiological detection indicators of modern medicine, including multi-modal data such as electrocardiogram signals, fingertip photoplethysmogram, and wrist pulse waveforms, and then uses these characteristic parameters to train a neural network model to find the potential relationship between the key parameters of physiological detection data and different stages of the development of traditional Chinese medicine courses, including deficiency, cold, dampness, stagnation, and stasis.

[0048] Make traditional Chinese medicine diagnosis objective, quantifiable, and digital.

[0049] Refer to Figure 1 , in some embodiments, the method for training a traditional Chinese medicine model includes:

[0050] S100. Obtain clinical data, where the clinical data includes: electrocardiogram signals, fingertip photoplethysmogram signals, and wrist pulse waveforms.

[0051] S200. Obtain the diagnostic results corresponding to the clinical data.

[0052] The collection of clinical data can be carried out in a hospital. When a traditional Chinese medicine doctor determines deficiency, cold, dampness, stagnation, and stasis in a patient, the electrocardiogram signal, fingertip photoplethysmogram signal, and wrist pulse waveform of the patient are collected; then the electrocardiogram signal, fingertip photoplethysmogram signal, wrist pulse waveform, and the doctor's diagnostic results are stored together.

[0053] Exemplarily, the waveform of the electrocardiogram signal is as Figure 2 shown, the waveform of the fingertip photoplethysmogram signal is as Figure 3 shown, and the wrist pulse waveform is as Figure 4 shown. Figure 2 . 3 In 4, the abscissa represents time and the ordinate represents amplitude. In practical applications, after obtaining the signals, we can first preprocess the electrocardiogram signal, fingertip photoplethysmogram signal, and wrist pulse waveform, such as noise reduction processing, low-pass filtering processing, and high-pass filtering processing.

[0054] S300. Use the diagnostic results as labels for the clinical data, and generate training data based on the clinical data and labels.

[0055] It should be noted that the electrocardiogram signal waveform shown in Figure 2 , the fingertip photoplethysmogram signal waveform shown in Figure 3 , and the wrist pulse waveform shown in Figure 4 can be directly used as training data. However, in order to reduce the demand for training data, solve the problem of reducing the data acquisition cost required for training, and obtain a machine learning model with satisfactory accuracy. Based on this, we can first extract the characteristic parameters of the electrocardiogram signal, fingertip photoplethysmogram signal, and wrist pulse waveform. The specific extraction method is not limited here. Exemplarily, the extraction method mentioned later can be used for extraction.

[0056] In this application, the traditional Chinese medicine model can be a classification model. According to the clinical data to be diagnosed output, the probability of the classification result is output, and then the result with the highest probability is selected as the output result of the model. Then the labels can be set as: deficiency, cold, dampness, stagnation, stasis, and others (such as non-deficiency, non-cold, non-dampness, non-stagnation, non-stasis).

[0057] S400. Use the training data to train the traditional Chinese medicine model to obtain a pre-trained traditional Chinese medicine model.

[0058] The training data is divided into training data, validation data, and test data according to a preset ratio (e.g., 7:2:1), and then the model can be trained. Among them, the training data can be used for model training, the validation data can be used to validate the model after each batch of training, and the test data can be used to test the model after the model training is completed to confirm the accuracy of the model.

[0059] In this application, by using electrocardiogram signals, fingertip photoplethysmograms, wrist pulse waveforms, and the diagnosis results of traditional Chinese medicine doctors, training data is generated, and then a traditional Chinese medicine model is trained to find the potential relationships between the key parameters of physiological detection data and different stages of the development of traditional Chinese medicine courses, including deficiency, cold, dampness, stagnation, and stasis, so as to make traditional Chinese medicine diagnosis objective, quantifiable, and digital.

[0060] Exemplarily, in the embodiments of this application, the processing process of multimodal physiological signals may include: signal preprocessing and feature parameter extraction.

[0061] Signal preprocessing

[0062] For the signal preprocessing of EEG signals, we can filter the EEG signals to eliminate noise. In this application, the band-pass frequency range of the filter can be set between 0 - 60 hz; exemplarily, it can be set to 0.5 Hz to 40 Hz; that is:

[0063]

[0064] In the above formula, H bandpass (f) represents the band-pass frequency of the filter, f L = 0.5 Hz (high-pass cut-off frequency), f H = 40 Hz (low-pass cut-off frequency), and j is the imaginary unit.

[0065] For the signal preprocessing of PPG signals, this application can use wavelet thresholding to denoise the PPG signals, that is:

[0066]

[0067] In the above formula, Ws,k represents the k-th wavelet coefficient at scale s, λ represents the adaptive threshold ( σ is the noise standard deviation), ψs,k(t) represents the wavelet basis function. Exemplarily, the wavelet basis function can be db4 wavelet, haar wavelet, morlet wavelet, etc.

[0068] In some embodiments, the diagnosis result consists of at least one of deficiency, cold, dampness, stagnation, and stasis.

[0069] Exemplarily, the diagnostic result is used as a label. Different scenarios require different diagnostic directions, so at least one of deficiency, cold, dampness, stasis, and blood stasis can be selected. In the present application, deficiency, cold, dampness, stasis, and blood stasis can be used as labels simultaneously.

[0070] In some embodiments, generating the training data according to the clinical data and the label includes:

[0071] Performing noise reduction and band-pass filtering on the electrocardiogram signal, that is, preprocessing the electrocardiogram signal.

[0072] Calculating the number of R waves in the electrocardiogram signal for a preset duration to obtain a heart rate parameter; wherein, the preset duration is greater than or equal to 30 seconds and less than or equal to 2 minutes;

[0073] Using the heart rate parameter and the first diagnostic result as the training data, wherein the first diagnostic result includes: deficiency and non-deficiency.

[0074] It should be noted that in traditional Chinese medicine diagnosis, "deficiency" refers to insufficient vital energy of the human body, decline of zang-fu organ functions, and decrease of disease resistance. Patients with deficiency syndrome often present symptoms such as pale complexion, listlessness, dizziness, fatigue, palpitations, and shortness of breath. In this embodiment, the method for extracting R waves is not limited. For example, a rule-based decision method and a deep learning-based detection method can be used.

[0075] In this embodiment, by calculating the number of R waves for a preset duration, taking 1 minute as the preset duration, the heart rate can be obtained. Then, using the heart rate as a feature parameter and the first diagnostic result as a label, a set of training data can be generated. The label includes deficiency and non-deficiency, and the model can obtain the result through binary classification.

[0076] Further, generating the training data according to the clinical data and the label includes: calculating the variation law of the R wave amplitude of the electrocardiogram signal to obtain a respiratory parameter;

[0077] Using the heart rate parameter and the first diagnostic result as the training data includes: using the heart rate parameter, the respiratory parameter, and the first diagnostic result as the training data.

[0078] In this embodiment, after extracting the R wave, the respiratory parameter (such as respiratory rate) can also be obtained according to the variation law of the R wave amplitude. The algorithm for obtaining the respiratory parameter based on the variation law of the R wave amplitude is not limited here. For example, the respiratory parameter can be calculated according to the statistic of the R wave amplitude.

[0079] Further, using the respiratory parameter and the heart rate as feature parameters and the first diagnostic result as a label, a set of training data can be generated.

[0080] In some embodiments, generating training data according to the clinical data and labels includes:

[0081] Obtaining the front and back N effective pulse wave amplitudes adjacent to the maximum amplitude waveform of the wrist pulse waveform; wherein, N is greater than or equal to 8 and less than or equal to 20;

[0082] Using the N effective pulse wave amplitudes and the second diagnosis result as training data, wherein the second diagnosis result includes: cold and non-cold.

[0083] "Cold" refers to the human body being invaded by cold pathogens, resulting in insufficient yang qi in the body and a slowdown in metabolism. Patients with cold syndromes often present symptoms such as aversion to cold, cold extremities, abdominal pain and diarrhea, and soreness of the waist and knees.

[0084] Regarding the definition of cold, in this embodiment, the front and back N effective pulse wave amplitudes adjacent to the maximum amplitude waveform of the wrist pulse waveform are used as training data, and the second diagnosis result is used as a label, thus generating a set of training data. The labels include cold and non-cold. Through binary classification by the model, the result can be obtained. That is, combined with the previous embodiment, only 2 binary classification models are required for the diagnosis model to achieve the goal.

[0085] In some embodiments, generating training data according to the clinical data and labels includes:

[0086] Obtaining TP, LF, HF, SDNN, SDANN, and RMSSD of heart rate variability according to the fingertip photoelectric pulse wave signal;

[0087] Using the TP, LF, HF, SDNN, SDANN, RMSSD, and the third diagnosis result as training data, wherein the third diagnosis result includes: damp and non-damp.

[0088] "Damp" refers to excessive damp pathogen in the human body, resulting in stagnation of water dampness and affecting normal physiological functions. Patients with damp syndromes often present symptoms such as heaviness and discomfort of the head and body, weakness of the limbs, abdominal distension and diarrhea, and tastelessness in the mouth.

[0089] Regarding the calculation of heart rate variability indexes (TP, LF, HF, SDNN, SDANN, RMSSD), we can perform frequency domain analysis of HRV through the following formula:

[0090]

[0091] In the above formula, RRn represents the period sequence of RR waves, with the unit of seconds; N represents the number of sampling points, and exemplarily N can be greater than or equal to 256; f represents the frequency, with the unit of Hz.

[0092] In this embodiment, TP, LF, and HF are frequency-domain information, and their specific meanings are as follows: TP: total power; LF: low-frequency power; HF: high-frequency power;

[0093] SDNN, SDANN, and RMSSD are time-domain information, and their specific meanings are as follows: SDNN: standard deviation of all normal sinus cardiac cycle intervals (NN), unit: ms. SDANN: The whole process is divided into continuous time periods of 5 minutes. First, calculate the standard deviation of the average value of each 5 minutes, and then calculate the average value of all standard deviations. Unit: ms. RMSSD: root mean square value of the difference between adjacent NN intervals throughout the whole process, unit: ms.

[0094] For "wet", in this embodiment, by obtaining the fingertip photoelectric pulse wave signal to obtain the frequency-domain information and time-domain information of heart rate variability as training data, and the third diagnosis result as a label, a set of training data can be generated. The labels include wet and non-wet. Through binary classification by the model, the result can be obtained, that is, combined with the above two embodiments, the diagnostic model only needs 3 binary classification models to achieve.

[0095] In some embodiments, the generating training data according to the clinical data and labels includes:

[0096] Obtain the time-frequency diagram corresponding to the fingertip photoelectric pulse wave signal;

[0097] Send the time-frequency diagram into a pre-trained convolutional neural network, and the pre-trained convolutional neural network outputs the peripheral microcirculation blockage parameters;

[0098] Use the peripheral microcirculation blockage parameters and the fourth diagnosis result as training data, where the fourth diagnosis result includes: coagulation and non-coagulation.

[0099] "Coagulation" means that the qi and blood in the human body do not circulate smoothly, resulting in blood stasis and the formation of blood stasis. Patients with coagulation syndrome often show symptoms such as pain, lumps, and purpura.

[0100] For "coagulation", in this embodiment, by obtaining the peripheral microcirculation blockage parameters as training data and the fourth diagnosis result as a label, a set of training data can be generated. The labels include coagulation and non-coagulation. Through binary classification by the model, the result can be obtained, that is, combined with the above three embodiments, the diagnostic model only needs 4 binary classification models to achieve.

[0101] In some embodiments, the generating training data according to the clinical data and labels includes:

[0102] Obtain the time difference between the R wave peaks of the electrocardiogram signal and the wave peaks of the fingertip photoelectric pulse wave signal;

[0103] Calculate the blood flow propagation speed parameter according to the time difference, the propagation speed of the electrocardiogram signal, and the propagation speed of the fingertip photoelectric pulse wave signal;

[0104] Use the blood flow propagation speed parameter and the fifth diagnostic result as training data, where the fifth diagnostic result includes: stasis and non-stasis.

[0105] "Stasis" refers to the unsmooth flow of qi and blood in the human body, resulting in the blockage of meridians and the formation of masses. Patients with stasis often experience symptoms such as pain, swelling, and numbness.

[0106] At the same time, use the fingertip photoelectric pulse wave signal and the electrocardiogram signal. Since the electrocardiogram signal propagates at the speed of light, while the fingertip photoelectric pulse is blood flow propagation. Analyze and find the peak of the fingertip photoelectric pulse and the R wave peak of the electrocardiogram signal respectively, and analyze the time difference between the two peaks to calculate the blood flow propagation time speed parameter.

[0107] Specifically, in one embodiment, a PWTT calculation model can be constructed as follows:

[0108] PWTT = argmax(cross_corr(ECG R-peake ,PPG foot ))

[0109] where ECG R-peake represents the R wave peak sequence of the electrocardiogram; PPG foot represents the valley time sequence of the photoplethysmogram.

[0110] In one embodiment, the blood flow propagation speed parameter of the present application can be calculated by the following formula.

[0111]

[0112] In the above formula, v represents the blood flow velocity, with the unit of cm / s; L represents the length of the aorta, which can be estimated by height. For example, L = 0.61×(h - 15.7); where h represents the user's height.

[0113] For "stasis", in this embodiment, by obtaining the propagation time speed parameter as training data and the fifth diagnostic result as a label, a set of training data can be generated. The labels include stasis and non-stasis. Through binary classification by the model, the result can be obtained, that is, combined with the above four embodiments, the diagnostic model only needs 5 binary classification models to achieve.

[0114] It should be noted that the above embodiments give the diagnosis of deficiency, cold, dampness, coagulation, and stasis using 5 binary classification models. However, it should be noted that in actual applications, we can use a six-classification model to achieve, and the classification results are: deficiency, cold, dampness, coagulation, stasis, and others.

[0115] Preferably, five binary classification models are used in this application to realize the diagnosis of deficiency, cold, dampness, stasis, and blood stasis, which can effectively reduce the number of model parameters, improve the convergence speed of the model, and then train a model that meets the requirements with less data.

[0116] Exemplarily, the TCM model architecture of this application is introduced. The TCM model can be a multi-modal feature fusion network, and the model architecture can be represented by the following formula:

[0117]

[0118] Wherein, represents the hidden state (dimension 128) at time step t; represents the ECG feature vector (dimension 32); represents h t-1 PPG feature vector (dimension 32).

[0119] Time-frequency feature extraction:

[0120]

[0121] In the above formula, represents the continuous wavelet transform time-frequency diagram of the PPG signal (size 128×128); W cwt represents the wavelet transform coefficient matrix, W c represents the convolutional layer weight matrix (3×3 convolution kernel)

[0122] Loss function design

[0123]

[0124] Wherein, p c represents the predicted probability distribution, y c represents the true label (one-hot encoding), θ represents the set of model parameters, and λ represents the L2 regularization coefficient (default 0.01).

[0125] In some embodiments, for the extraction of the feature parameters of this application, dynamic time warping (DWT) is used, specifically as follows:

[0126]

[0127] Wherein, P and Q represent the pulse wave sequences to be aligned, and W represents the optimal warping path (meeting the monotonicity and continuity constraints).

[0128] Sample entropy calculation

[0129]

[0130] Among them, m represents the embedding dimension, with a default value of 2, r represents the tolerance, usually taking 0.2 times the standard deviation, N represents the data length. Exemplarily, N is greater than or equal to 1000.

[0131] Exemplarily, the present application can use the following model optimization techniques

[0132] Attention mechanism

[0133]

[0134] a i represents the attention weight at time step i, v, W h represent trainable parameter matrices, h i represents the LSTM hidden state (dimension 128).

[0135] Transfer learning strategy

[0136]

[0137] θ target represents the pre-trained parameters in the source domain, γ represents the transfer intensity coefficient (range 0.1 - 1.0), D target represents the target domain dataset.

[0138] In some embodiments, the evaluation metrics of the model can include: model accuracy, macro-average F1, and AUC-ROC curve; the calculation formulas for these three metrics are given below:

[0139] Model accuracy: Among them, TP is the number of positive samples that are correctly identified, where the predicted result is a positive sample and the actual result is also a positive sample. FP is the number of false-negative samples that are misreported, where the predicted result is a positive sample but the actual result is a negative sample. TN is the number of negative samples that are correctly identified, where the predicted result is a negative sample and the actual result is also a negative sample. FN is the number of positive samples that are missed, where the predicted result is a negative sample but the actual result is a positive sample.

[0140] Macro-average F1: Among them, P c represents the precision rate of class c; R c represents the recall rate of class c.

[0141] AUC-ROC curve: Among them, TPR is the true positive rate, and FPR is the false positive rate.

[0142] SHAP interpretability analysis:

[0143]

[0144] The SHAP value (contribution degree) representing feature i, S represents the feature subset, M represents the total number of features (in this solution, M = 32), and f(·) represents the model prediction function.

[0145] Example 2

[0146] Referring to Figure 5 In a second aspect, the present application provides a method for applying a traditional Chinese medicine model, and the method for applying the traditional Chinese medicine model includes:

[0147] S500. Guide the user through an interaction device to collect the clinical data to be diagnosed of the user;

[0148] S600. Send the clinical data to be diagnosed into the traditional Chinese medicine model trained by the above-mentioned traditional Chinese medicine model training method, and the traditional Chinese medicine model outputs a diagnosis result; and,

[0149] S700. Display the diagnosis result on the interaction interface.

[0150] In this way, after diagnosis by a traditional Chinese medicine doctor, the diagnosis result displayed on the interaction interface can be used as a reference to improve the consistency of diagnosis.

[0151] Referring to Figure 1 In some embodiments, the traditional Chinese medicine model training method includes:

[0152] S100. Obtain clinical data, where the clinical data includes: electrocardiogram signal, fingertip photoelectric pulse wave signal, and wrist pulse waveform;

[0153] S200. Obtain the diagnosis result corresponding to the clinical data;

[0154] S300. Use the diagnosis result as the label of the clinical data, and generate training data according to the clinical data and the label;

[0155] S400. Use the training data to train the traditional Chinese medicine model to obtain a pre-trained traditional Chinese medicine model.

[0156] In some embodiments, the diagnosis result consists of at least one of deficiency, cold, dampness, stasis, and congestion.

[0157] In some embodiments, the generating training data according to the clinical data and the label includes:

[0158] Perform noise reduction and band-pass filtering on the electrocardiogram signal;

[0159] Calculate the number of R waves of the electrocardiogram signal within a preset duration to obtain a heart rate parameter; wherein, the preset duration is greater than or equal to 30 seconds and less than or equal to 2 minutes;

[0160] Use the heart rate parameter and the first diagnosis result as training data, where the first diagnosis result includes: deficient and non-deficient.

[0161] In some embodiments, generating training data according to the clinical data and labels includes: calculating the variation law of the R-wave amplitude of the electrocardiogram signal to obtain a respiratory parameter;

[0162] Using the heart rate parameter and the first diagnosis result as training data includes: using the heart rate parameter, the respiratory parameter, and the first diagnosis result as training data.

[0163] In some embodiments, generating training data according to the clinical data and labels includes:

[0164] Obtain the N effective pulse wave amplitudes before and after adjacent to the maximum amplitude waveform of the wrist pulse waveform; where N is greater than or equal to 8 and less than or equal to 20;

[0165] Use the N effective pulse wave amplitudes and the second diagnosis result as training data, where the second diagnosis result includes: cold and non-cold.

[0166] In some embodiments, generating training data according to the clinical data and labels includes:

[0167] Obtain TP, LF, HF, SDNN, SDANN, RMSSD of heart rate variability according to the fingertip photoelectric pulse wave signal;

[0168] Use the TP, LF, HF, SDNN, SDANN, RMSSD and the third diagnosis result as training data, where the third diagnosis result includes: damp and non-damp.

[0169] In some embodiments, generating training data according to the clinical data and labels includes:

[0170] Obtain the time-frequency diagram corresponding to the fingertip photoelectric pulse wave signal;

[0171] Send the time-frequency diagram into a pre-trained convolutional neural network, and the pre-trained convolutional neural network outputs the peripheral microcirculation blockage parameter;

[0172] Use the peripheral microcirculation blockage parameter and the fourth diagnosis result as training data, where the fourth diagnosis result includes: coagulation and non-coagulation.

[0173] In some embodiments, generating training data according to the clinical data and labels includes:

[0174] Obtain the time difference between the R-wave peak of the electrocardiogram signal and the peak of the fingertip photoelectric pulse wave signal;

[0175] Calculate the blood flow propagation speed parameter based on the time difference, the propagation speed of the electrocardiogram signal, and the propagation speed of the fingertip photoplethysmogram signal;

[0176] Use the blood flow propagation speed parameter and the fifth diagnosis result as training data, where the fifth diagnosis result includes: stasis and non-stasis.

[0177] Example 3

[0178] In a third aspect, the present application provides a traditional Chinese medicine diagnosis device, which includes:

[0179] A traditional Chinese medicine diagnosis device, which includes:

[0180] A clinical data collection device to be diagnosed, used to collect clinical data to be diagnosed;

[0181] A memory that stores a traditional Chinese medicine model; the traditional Chinese medicine model is trained using the traditional Chinese medicine model training method described above.

[0182] A processor, connected to the memory, and the processor is configured to call the traditional Chinese medicine model to output a diagnosis result based on the clinical data to be diagnosed.

[0183] In an embodiment, the traditional Chinese medicine diagnosis device further includes:

[0184] An interaction device, connected to the processor, and the interaction device is used to guide the user to collect the clinical data to be diagnosed by the user; and display the diagnosis result.

[0185] In some embodiments, the memory stores a traditional Chinese medicine model application; the processor is configured to execute the traditional Chinese medicine model application program to implement the above-mentioned traditional Chinese medicine model application method.

[0186] It should be noted that in the above embodiments, the descriptions of the various embodiments have their own emphases. For the parts not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0187] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0188] The present invention will be described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one flow Figure 1 one flow or more flows and / or blocks Figure 1 one block or more blocks.

[0189] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in one flow Figure 1 one flow or more flows and / or blocks Figure 1 one block or more blocks.

[0190] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operational steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or more flows and / or blocks Figure 1 one block or more blocks.

[0191] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0192] Obviously, those skilled in the art can make various changes and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and variations.

Claims

1. A traditional Chinese medicine model training method, characterized in that Including: Obtaining clinical data, where the clinical data includes: electrocardiogram signals, fingertip photoplethysmogram signals, and wrist pulse waveforms; Obtaining the diagnosis result corresponding to the clinical data; Using the diagnosis result as a label for the clinical data, and generating training data based on the clinical data and the label; Training a traditional Chinese medicine model using the training data to obtain a pre-trained traditional Chinese medicine model.

2. The method for training a traditional Chinese medicine model according to claim 1, wherein The diagnosis result is composed of at least one of deficiency, cold, dampness, stasis, and congestion.

3. The traditional Chinese medicine model training method according to claim 2, characterized in that, The generating training data based on the clinical data and the label includes: Performing noise reduction and band-pass filtering on the electrocardiogram signals; Calculating the number of R waves in the electrocardiogram signals for a preset duration to obtain a heart rate parameter; wherein the preset duration is greater than or equal to 30 seconds and less than or equal to 2 minutes; Using the heart rate parameter and the first diagnosis result as training data, where the first diagnosis result includes: deficiency and non-deficiency.

4. The method for training a traditional Chinese medicine model according to claim 3, wherein The generating training data based on the clinical data and the label includes: calculating the variation law of the R wave amplitude of the electrocardiogram signals to obtain a respiratory parameter; The using the heart rate parameter and the first diagnosis result as training data includes: using the heart rate parameter, the respiratory parameter, and the first diagnosis result as training data.

5. The traditional Chinese medicine model training method according to claim 3 or 4, characterized in that The generating training data based on the clinical data and the label includes: Obtaining the front and back N effective pulse wave amplitudes adjacent to the maximum amplitude waveform of the wrist pulse waveform; wherein N is greater than or equal to 8 and less than or equal to 20; Using the N effective pulse wave amplitudes and the second diagnosis result as training data, where the second diagnosis result includes: cold and non-cold.

6. The traditional Chinese medicine model training method according to claim 3 or 4, characterized in that, The generating training data based on the clinical data and the label includes: Obtaining TP, LF, HF, SDNN, SDANN, and RMSSD of heart rate variability according to the fingertip photoplethysmogram signals; Using the TP, LF, HF, SDNN, SDANN, RMSSD, and the third diagnosis result as training data, where the third diagnosis result includes: dampness and non-dampness.

7. The traditional Chinese medicine model training method according to claim 6, wherein The generating training data based on the clinical data and the label includes: Obtaining the time-frequency diagram corresponding to the fingertip photoplethysmogram signals; Feeding the time-frequency diagram into a pre-trained convolutional neural network, and the pre-trained convolutional neural network outputs the parameters of peripheral microcirculation blockage; Using the parameters of peripheral microcirculation blockage and the fourth diagnosis result as training data, where the fourth diagnosis result includes: stasis and non-stasis.

8. The traditional Chinese medicine model training method according to claim 6, wherein, The generating training data based on the clinical data and the label includes: Obtaining the time difference between the R wave peaks of the electrocardiogram signals and the wave peaks of the fingertip photoplethysmogram signals; Calculating the blood flow propagation speed parameter according to the time difference, the propagation speed of the electrocardiogram signals, and the propagation speed of the fingertip photoplethysmogram signals; Using the blood flow propagation speed parameter and the fifth diagnosis result as training data, where the fifth diagnosis result includes: congestion and non-congestion.

9. A traditional Chinese medicine diagnosis device, characterized in that, Including: A clinical data acquisition device to be diagnosed, used for collecting clinical data to be diagnosed; A memory that stores a traditional Chinese medicine model; the traditional Chinese medicine model is obtained by training using the traditional Chinese medicine model training method according to any one of claims 1-8. A processor connected to the memory, the processor being configured to call the traditional Chinese medicine model to output a diagnosis result based on the clinical data to be diagnosed.

10. The traditional Chinese medicine diagnosis device according to claim 9, characterized in that, It includes: An interaction device connected to the processor, the interaction device being used to guide a user to collect the clinical data to be diagnosed by the user; And display the diagnosis result.

Images