Intelligent traditional Chinese medicine pulse diagnosis system

Abstract

The invention discloses an intelligent traditional Chinese medicine pulse diagnosis system which comprises a pulse condition information acquisition module (1), a processing module (2), a diagnosis module (3) and a visualization module (4). According to the method, the adaptive Fourier algorithm AFD and wavelet transformation are combined, complex pulse signals can be decomposed more accurately, key physiological features are extracted, the pulse diagnosis signal processing precision and reliability are improved, the key features in the pulse signals can be accurately extracted by means of the adaptive Fourier decomposition algorithm, and the pulse diagnosis signal processing accuracy and reliability are improved. High accuracy and consistency of pulse condition type classification are ensured, multiple common traditional Chinese medicine pulse condition types can be accurately identified through the deep learning model, diversified and complex pulse condition types are coped, an objective and accurate basis is provided for traditional Chinese medicine diagnosis, subjectivity and uncertainty of traditional Chinese medicine diagnosis can be reduced, and diagnosis accuracy is improved. The accuracy and reliability of diagnosis are improved, and the accuracy of pulse diagnosis is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, in particular to an intelligent traditional Chinese medicine pulse diagnosis system. Background Art

[0002] Pulse diagnosis is one of the four diagnostic methods in traditional Chinese medicine. It is a unique diagnostic method that uses the feeling of the fingers to analyze the "position, number, shape, and momentum" of the pulse to judge the functional state of the internal organs, the rise and fall of qi and blood, and the trend of changes in evil and righteousness. Therefore, pulse diagnosis can determine the cause of the disease, know the location of the disease, examine the pathogenesis, distinguish the nature of the disease, and observe the prognosis, thereby achieving the purpose of non-invasive diagnosis. It has a positive significance for the diagnosis and treatment of diseases. Traditional Chinese medicine pulse diagnosis relies on the doctor's finger touch, and judges the condition by sensing the strength, speed, and shape of the pulse. However, this method is highly dependent on the doctor's experience and technique, and is easily affected by In order to overcome the influence of subjective factors, an intelligent Chinese medicine pulse diagnosis system is proposed to collect and analyze the position, number, shape and potential characteristics of the pulse, and finally intelligently analyze the single pulse and combined pulse categories and dozens of pulse parameters in the time-frequency-domain and output a standard pulse diagram. At the same time, it can record and track the changes in pulse characteristics in different periods, which has important reference value for the evaluation of the efficacy of the disease and provides an objective basis for the identification of health status and the evaluation of intervention effects. The existing intelligent Chinese medicine pulse diagnosis system cannot handle complex pulse diagnosis data characteristics and has difficulty coping with diverse and complex pulse types, which affects the accuracy and reliability of diagnosis. Summary of the invention

[0003] The purpose of the present invention is to provide an intelligent Chinese medicine pulse diagnosis system to solve the problems raised in the above background technology.

[0004] To achieve the above object, the present invention provides the following technical solutions: an intelligent Chinese medicine pulse diagnosis system, comprising 1) a pulse information acquisition module, 2) a processing module, wherein the processing module includes a diagnosis database, 3) a diagnosis module, and 4) a visualization module;

[0005] 1) The pulse information acquisition module uses an FMCW radar sensor and a mixer to detect the physiological signals of each patient respectively, and converts the physiological signals into pulse data corresponding to the patient, obtains the pulse data of the patient, and transmits the pulse data to the processing module;

[0006] 2) The processing module receives the pulse data transmitted by the FMCW radar sensor, performs signal processing on the pulse data, and transmits the processed pulse data to the diagnostic database for storage;

[0007] 3) the diagnosis module performs intelligent feature analysis on the pulse data in the diagnosis database, and the diagnosis module is connected to the FMCW radar sensor via a wire;

[0008] 4) The visualization module has a pre-trained deep learning model built in, which can identify a variety of common TCM pulse types, generate characteristic diagnostic strategies, and display the characteristic diagnostic strategies;

[0009] The specific steps of the intelligent Chinese medicine pulse diagnosis system are:

[0010] S1. The FMCW radar sensor transmits a series of continuous frequency modulated millimeter waves. The form of the transmitted signal is expressed as:

[0011]

[0012] Where A is the amplitude, f o is the center frequency, k is the slope of the frequency change, is the initial phase,

[0013] When the transmitted signal encounters the wrist skin, it will be reflected back to form a reflected signal. The reflected signal will have a certain delay τ and attenuation a during the transmission process, so the received reflected signal is in the form of:

[0014] x Rx (t) = ax t (t-τ)

[0015] The mixer mixes the received reflected signal with the transmitted signal. The output of the mixer is the difference in instantaneous frequencies of the two input signals, that is, x IF (t) = x t2 (t)×x R2 (t)

[0016] After mathematical derivation and low-pass filtering, the intermediate frequency signal can be obtained:

[0017] x IF (t) = A″cos(2πKτt+2πf o τ+πKτ 2 )

[0018] The intermediate frequency signal contains the frequency and displacement information related to the pulse. By accurately measuring the frequency and phase changes of the intermediate frequency signal, high-precision pulse data is obtained and transmitted to the processing module for signal processing;

[0019] S2, the processing module pre-processes the pulse signal through Kalman filtering and adaptive filtering technology, first uses adaptive filtering to perform preliminary denoising on the pulse signal, and then uses Kalman filtering to dynamically estimate and predict the signal to further suppress noise interference and improve the purity of the signal, ensuring that the extracted pulse feature signal is pure;

[0020] S3. The pulse signal after filtering will be decomposed by the time-frequency joint analysis method in the diagnosis module. The time-frequency joint analysis method combines the advantages of time domain and frequency domain analysis, and can provide information on the pulse wave in both time and frequency dimensions. The time-frequency joint analysis method uses the adaptive Fourier algorithm AFD and wavelet transform technology to decompose the pulse wave into local features of different time and frequency, so as to more accurately reflect the dynamic changes of the pulse.

[0021] The adaptive Fourier algorithm can dynamically adjust the decomposition strategy according to the changes in the pulse signal, so as to more accurately capture the subtle changes in the pulse. Through the matching pursuit method, the optimal basis combination is adaptively selected according to the decomposition signal. The decomposition process can be expressed as:

[0022]

[0023] Among them, A n is the decomposition coefficient, B n is the basis function, and the AFD method uses the Blaschke product to construct its basis function:

[0024]

[0025] According to the maximum selection principle, the energy is extracted to the maximum extent, and appropriate basis functions are selected to approximate the original function, so as to achieve rapid energy convergence and provide accurate and data-based diagnostic results.

[0026] The time-frequency characteristics of the pulse signal are extracted by wavelet transform, and the signal is convolved with the wavelet basis function. The convolution operation is performed on the pulse signal x(t) and the wavelet basis function ψ(t). The result W(a,b) is a new function. The wavelet basis function ψ(t) is transformed into ψ after translation and scaling. a,b (t), the specific form is:

[0027]

[0028] Among them, a is the scale factor, which determines the degree of expansion and contraction of the wavelet basis function, that is, the frequency characteristic; b is the time shift factor, which determines the position of the wavelet basis function on the time axis;

[0029] The formula for the convolution operation is:

[0030]

[0031] The formula represents the combination of the pulse signal x(t) and the wavelet basis function ψ a,b (t) Integrate over the entire time domain, and the result W(a,b) is a complex number that contains the characteristic information of the signal at scale a and time position b;

[0032] By adjusting the values ​​of the scale factor a and the time shift factor b, the time-frequency characteristics of the signal in different frequency bands and time periods can be obtained. The time-frequency characteristics include energy spectrum density, instantaneous frequency, and instantaneous amplitude. The time-frequency characteristics can reflect the changing characteristics of the pulse signal at different times and frequencies.

[0033] S4. The extracted pulse signal feature data is input into the deep learning model built into the visualization module. The deep learning model can learn the features of different pulse types by learning a large amount of pulse diagnosis data, and accurately classify the pulse types, such as floating pulse, sinking pulse, slippery pulse, and stringy pulse. The pulse waveform and feature parameters are intuitively displayed in a graphical form, and a diagnosis report can be automatically generated to help doctors make decisions.

[0034] Preferably, the specific steps of pre-training the deep learning model are:

[0035] (1) Collect a large amount of pulse signal feature data. Pulse diagnosis data includes the patient's pulse characteristics, such as the pulse beat, strength, and frequency. After collecting the data, preprocessing is required, including data cleaning and normalization, to improve the training efficiency and classification accuracy of the model. The preprocessed pulse signal feature data is divided into a training set and a test set;

[0036] (2) Select a suitable deep learning model architecture, such as convolutional neural network (CNN), and use the training set as input to perform model training. During the training process, the model parameters are adjusted and the optimization algorithm is optimized so that the model can learn the characteristics of different TCM pulse types;

[0037] (3) Use the test set to evaluate the trained deep learning model and calculate the model's accuracy, recall and other indicators. Based on the evaluation results, further optimize and improve the model to improve the model's recognition performance and generalization ability.

[0038] Compared with the prior art, the beneficial effects of the present invention are as follows: contactless pulse signal acquisition is achieved through radar technology, and the tiny vibrations on the wrist surface caused by the pulse are accurately captured, which not only avoids the errors caused by pressure and posture changes in traditional contact pulse diagnosis, provides more stable and accurate pulse data, but also reduces the discomfort of patients during diagnosis and treatment. It is particularly suitable for groups that are sensitive to contact, such as infants, the elderly, and critically ill patients, and the contactless design also reduces the risk of cross-infection, especially during epidemics of infectious diseases; the processing module first uses adaptive filtering to perform preliminary denoising on the pulse signal, and then uses Kalman filtering to dynamically estimate and predict the signal to further suppress noise interference and improve the purity of the signal, ensuring the purity and accuracy of the signal, and maintaining the stability and accuracy of data processing in various complex environments, continuously outputting high-quality signal processing results, avoiding performance degradation, and ensuring the reliability of the diagnosis results;

[0039] The high-purity pulse signal is decomposed into local features of different time and frequency through the adaptive Fourier algorithm AFD and wavelet transform technology. It has higher resolution and adaptability, can more accurately capture the subtle changes in the pulse signal, and provide a reliable data basis for the subsequent pulse type classification. The combination of the adaptive Fourier algorithm AFD and wavelet transform can more accurately decompose the complex pulse signal and extract the key physiological features, thereby effectively solving the problem of being unable to process complex pulse diagnosis data features and improving the accuracy and reliability of pulse diagnosis signal processing. With the help of the adaptive Fourier decomposition algorithm, the system can accurately extract the key features in the pulse signal, ensure the high accuracy and consistency of pulse type classification, and help doctors diagnose diseases and assess patients' health more accurately.

[0040] By studying a large amount of pulse diagnosis data, the deep learning model can learn the characteristics of different pulse types. The deep learning model can accurately identify a variety of common TCM pulse types and cope with diverse and complex pulse types, providing an objective and accurate basis for TCM diagnosis. It not only improves the accuracy and reliability of pulse diagnosis, but also reduces diagnostic deviations caused by human factors, and promotes the standardization and digitization of pulse diagnosis technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a system flow chart of the present invention;

[0042] Figure 2 This is a flow chart of the training deep learning model of the present invention. DETAILED DESCRIPTION

[0043] The technical solution of the present invention is described in detail below through specific embodiments.

[0044] See also Figure 1-2 , the present invention provides an intelligent Chinese medicine pulse diagnosis system, including 1) a pulse information acquisition module, 2) a processing module, and the processing module includes a diagnosis database, 3) a diagnosis module, and 4) a visualization module;

[0045] 1) The pulse information acquisition module uses an FMCW radar sensor and a mixer to detect the physiological signals of each patient respectively, and converts the physiological signals into pulse data corresponding to the patient, obtains the pulse data of the patient, and transmits the pulse data to the processing module;

[0046] 2) The processing module receives the pulse data transmitted by the FMCW radar sensor, performs signal processing on the pulse data, and transmits the processed pulse data to the diagnostic database for storage;

[0047] 3) the diagnosis module performs intelligent feature analysis on the pulse data in the diagnosis database, and the diagnosis module is connected to the FMCW radar sensor via a wire;

[0048] 4) The visualization module has a pre-trained deep learning model built in. Through the deep learning model, it can identify a variety of common TCM pulse types, generate characteristic diagnostic strategies, and display the characteristic diagnostic strategies for medical staff to view and diagnose;

[0049] The specific steps of the intelligent Chinese medicine pulse diagnosis system are:

[0050] S1. The FMCW radar sensor transmits a series of continuous frequency modulated millimeter waves. The form of the transmitted signal is expressed as:

[0051]

[0052] Where A is the amplitude, f o is the center frequency, k is the slope of the frequency change, is the initial phase,

[0053] When the transmitted signal encounters the wrist skin, it will be reflected back to form a reflected signal. The reflected signal will have a certain delay τ and attenuation a during the transmission process, so the received reflected signal is in the form of:

[0054] x Rx (t) = ax t (t-τ)

[0055] The mixer mixes the received reflected signal with the transmitted signal. The output of the mixer is the difference in instantaneous frequencies of the two input signals, that is, x IF (t) = x t2 (t)×x R2 (t)

[0056] After mathematical derivation and low-pass filtering, the intermediate frequency signal can be obtained:

[0057] x IF (t) = A″cos(2πKτt+2πf o τ+πKτ 2 )

[0058] The intermediate frequency signal contains the frequency and displacement information related to the pulse. By accurately measuring the frequency and phase changes of the intermediate frequency signal, high-precision pulse data is obtained and transmitted to the processing module for signal processing;

[0059] S2, the processing module pre-processes the pulse signal through Kalman filtering and adaptive filtering technology, first uses adaptive filtering to perform preliminary denoising on the pulse signal, and then uses Kalman filtering to dynamically estimate and predict the signal to further suppress noise interference and improve the purity of the signal, ensuring that the extracted pulse feature signal is pure;

[0060] In pulse signal processing, adaptive filtering can be used to remove interference factors such as electromyographic noise, motion artifacts, and environmental noise. Adaptive filtering can be used in combination with Kalman filtering to improve the signal-to-noise ratio and extraction accuracy of pulse signals. Kalman filtering can detect the motion of the measured object in real time based on acceleration information, thereby dynamically adjusting the filtering parameters, realizing adaptive filtering based on motion measurement, eliminating motion noise interference, and realizing dynamic extraction of pulse rate. It can estimate the dynamic changes of pulse signals and effectively suppress noise interference.

[0061] S3. The pulse signal after filtering will be decomposed by the time-frequency joint analysis method in the diagnosis module. The time-frequency joint analysis method combines the advantages of time domain and frequency domain analysis, and can provide information on the pulse wave in both time and frequency dimensions. The time-frequency joint analysis method uses the adaptive Fourier algorithm AFD and wavelet transform technology to decompose the pulse wave into local features of different time and frequency, so as to more accurately reflect the dynamic changes of the pulse.

[0062] The adaptive Fourier algorithm is used to convert the time domain pulse signal into the frequency domain signal, decompose the pulse wave into components of different frequencies, and calculate the amplitude, phase and other parameters of each frequency component. By comparing the spectral characteristics of different pulse patterns, the relationship between the pulse pattern and physiological and pathological changes such as organ function and qi and blood status can be revealed.

[0063] The adaptive Fourier algorithm can dynamically adjust the decomposition strategy according to the changes in the pulse signal, so as to more accurately capture the subtle changes in the pulse. Through the matching pursuit method, the optimal basis combination is adaptively selected according to the decomposition signal. The decomposition process can be expressed as:

[0064]

[0065] Among them, A n is the decomposition coefficient, B n is the basis function, and the AFD method uses the Blaschke product to construct its basis function:

[0066]

[0067] According to the maximum selection principle, the energy is extracted to the maximum extent, and appropriate basis functions are selected to approximate the original function, so as to achieve rapid energy convergence and provide accurate and data-based diagnostic results.

[0068] The time-frequency characteristics of the pulse signal are extracted by wavelet transform, and the signal is convolved with the wavelet basis function. The convolution operation is performed on the pulse signal x(t) and the wavelet basis function ψ(t). The result W(a,b) is a new function. The wavelet basis function ψ(t) is transformed into ψ after translation and scaling. a,b (t), the specific form is:

[0069]

[0070] Among them, a is the scale factor, which determines the degree of expansion and contraction of the wavelet basis function, that is, the frequency characteristic; b is the time shift factor, which determines the position of the wavelet basis function on the time axis;

[0071] The formula for the convolution operation is:

[0072]

[0073] The formula represents the combination of the pulse signal x(t) and the wavelet basis function ψ a,b (t) Integrate over the entire time domain, and the result W(a,b) is a complex number that contains the characteristic information of the signal at scale a and time position b;

[0074] By adjusting the values ​​of the scale factor a and the time shift factor b, the time-frequency characteristics of the signal in different frequency bands and time periods can be obtained. The time-frequency characteristics include energy spectrum density, instantaneous frequency, and instantaneous amplitude. The time-frequency characteristics can reflect the changing characteristics of the pulse signal at different times and frequencies.

[0075] S4. The extracted pulse signal feature data is input into the deep learning model built into the visualization module. The specific steps of pre-training the deep learning model are:

[0076] (1) Collect a large amount of pulse signal feature data. Pulse diagnosis data includes the patient's pulse characteristics, such as the pulse beat, strength, and frequency. After collecting the data, preprocessing is required, including data cleaning and normalization, to improve the training efficiency and classification accuracy of the model. The preprocessed pulse signal feature data is divided into a training set and a test set;

[0077] (2) Select a suitable deep learning model architecture, such as convolutional neural network (CNN), and use the training set as input to perform model training. During the training process, the model parameters are adjusted and the optimization algorithm is optimized so that the model can learn the characteristics of different TCM pulse types;

[0078] (3) Use the test set to evaluate the trained deep learning model, calculate the model's accuracy, recall and other indicators, and further optimize and improve the model based on the evaluation results to improve the model's recognition performance and generalization ability;

[0079] By learning a large amount of pulse diagnosis data, the deep learning model can learn the characteristics of different pulse types and achieve accurate classification of pulse types, such as floating pulse, sinking pulse, slippery pulse, and stringy pulse. It can intuitively display the pulse waveform and characteristic parameters in a graphical form and automatically generate a diagnosis report. Medical staff can understand the patient's pulse characteristics and diagnosis results more intuitively, helping medical staff to obtain comprehensive information about the patient more quickly, improve their cognition and understanding of the disease, enhance clinical thinking and diagnostic capabilities, reduce the risk of missed diagnosis and misdiagnosis, and thus improve the accuracy and efficiency of diagnosis.

[0080] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An intelligent Chinese medicine pulse diagnosis system, characterized by: It includes a pulse information acquisition module, a processing module, a diagnosis module and a visualization module, and the processing module includes a diagnosis database; The pulse information acquisition module uses an FMCW radar sensor and a mixer to detect the physiological signals of each patient, convert the physiological signals into pulse data corresponding to the patient, obtain the pulse data of the patient, and transmit the pulse data to the processing module; The processing module receives the pulse data transmitted by the FMCW radar sensor, performs signal processing on the pulse data, and transmits the processed pulse data to the diagnostic database for storage; The diagnosis module performs intelligent feature analysis on the pulse data in the diagnosis database, and the diagnosis module is connected to the FMCW radar sensor via a wire; The visualization module has a pre-trained deep learning model built in. Through the deep learning model, it can identify a variety of common TCM pulse types, generate characteristic diagnosis strategies, and display the characteristic diagnosis strategies; The specific steps of the intelligent Chinese medicine pulse diagnosis system are: S1. The FMCW radar sensor transmits a series of continuous frequency modulated millimeter waves. The form of the transmitted signal is expressed as: Where A is the amplitude, f o is the center frequency, k is the slope of the frequency change, is the initial phase, When the transmitted signal encounters the wrist skin, it will be reflected back to form a reflected signal. The reflected signal will have a certain delay τ and attenuation a during the transmission process, so the received reflected signal is in the form of: x Rx (t)=ax t (t-τ) The mixer mixes the received reflected signal with the transmitted signal. The output of the mixer is the difference in instantaneous frequencies of the two input signals, that is, x IF (t) = x t2 (t)×x R2 (t) After mathematical derivation and low-pass filtering, the intermediate frequency signal can be obtained: x IF (t)=A″cos(2πKτt+2πf o τ+πKτ 2 ) The intermediate frequency signal contains the frequency and displacement information related to the pulse. By accurately measuring the frequency and phase changes of the intermediate frequency signal, high-precision pulse data is obtained and transmitted to the processing module for signal processing; S2, the processing module pre-processes the pulse signal through Kalman filtering and adaptive filtering technology, first uses adaptive filtering to perform preliminary denoising on the pulse signal, and then uses Kalman filtering to dynamically estimate and predict the signal to further suppress noise interference and improve the purity of the signal, ensuring that the extracted pulse feature signal is pure; S3. The pulse signal after filtering will be decomposed by the time-frequency joint analysis method in the diagnosis module. The time-frequency joint analysis method combines the advantages of time domain and frequency domain analysis, and can provide information on the pulse wave in both time and frequency dimensions. The time-frequency joint analysis method uses the adaptive Fourier algorithm AFD and wavelet transform technology to decompose the pulse wave into local features of different time and frequency, so as to more accurately reflect the dynamic changes of the pulse. The adaptive Fourier algorithm can dynamically adjust the decomposition strategy according to the changes in the pulse signal, so as to more accurately capture the subtle changes in the pulse. Through the matching pursuit method, the optimal basis combination is adaptively selected according to the decomposition signal. The decomposition process can be expressed as: Among them, A n is the decomposition coefficient, B n is the basis function, and the AFD method uses the Blaschke product to construct its basis function: According to the maximum selection principle, the energy is extracted to the maximum extent, and appropriate basis functions are selected to approximate the original function, so as to achieve rapid energy convergence and provide accurate and data-based diagnostic results. The time-frequency characteristics of the pulse signal are extracted by wavelet transform, and the signal is convolved with the wavelet basis function. The convolution operation is performed on the pulse signal x(t) and the wavelet basis function ψ(t). The result W(a,b) is a new function. The wavelet basis function ψ(t) is transformed into ψ after translation and scaling. a,b (t), the specific form is: Among them, a is the scale factor, which determines the degree of expansion and contraction of the wavelet basis function, that is, the frequency characteristic; b is the time shift factor, which determines the position of the wavelet basis function on the time axis; The formula for the convolution operation is: The formula represents the combination of the pulse signal x(t) and the wavelet basis function ψ a,b (t) Integrate over the entire time domain, and the result W(a,b) is a complex number that contains the characteristic information of the signal at scale a and time position b; By adjusting the values ​​of the scale factor a and the time shift factor b, the time-frequency characteristics of the signal in different frequency bands and time periods can be obtained. The time-frequency characteristics include energy spectrum density, instantaneous frequency, and instantaneous amplitude. The time-frequency characteristics can reflect the changing characteristics of the pulse signal at different times and frequencies. S4. The extracted pulse signal feature data is input into the deep learning model built into the visualization module. By learning a large amount of pulse diagnosis data, the deep learning model can learn the characteristics of different pulse types and accurately classify pulse types, such as floating pulse, sinking pulse, slippery pulse, and string pulse. The pulse waveform and feature parameters are intuitively displayed in a graphical form, and a diagnostic report can be automatically generated to help doctors make decisions.

2. An intelligent Chinese medicine pulse diagnosis system according to claim 1, characterized in that: Specific steps for pre-training deep learning models: (1) Collect a large amount of pulse signal feature data. Pulse diagnosis data includes the patient's pulse characteristics, such as the pulse beat, strength, and frequency. After collecting the data, preprocessing is required, including data cleaning and normalization, to improve the training efficiency and classification accuracy of the model. The preprocessed pulse signal feature data is divided into a training set and a test set; (2) Select a suitable deep learning model architecture, such as convolutional neural network (CNN), and use the training set as input to perform model training. During the training process, the model parameters are adjusted and the optimization algorithm is optimized so that the model can learn the characteristics of different TCM pulse types; (3) Use the test set to evaluate the trained deep learning model and calculate the model's accuracy, recall and other indicators. Based on the evaluation results, further optimize and improve the model to improve the model's recognition performance and generalization ability.

Images