Acupuncture control optimization system based on fNIRS-guided attention neural network

Abstract

The present invention discloses an attention neural network acupuncture control optimization system based on fNIRS guidance, including: an acquisition module, which is configured to: acquire physiological signals of the brain of the subject during acupuncture; a preprocessing module, which is configured to: preprocess the acquired physiological signals to obtain signal characteristics of each brain region; a prediction module, which is configured to: input the signal characteristics of each brain region into the trained attention neural network, and output the quantitative prediction results of acupuncture efficacy of each brain region; a control module, which is configured to: control the duration and position of acupuncture according to the quantitative prediction results of acupuncture efficacy of each brain region. Computer technology is used to assist acupuncture workers in finding more accurate acupuncture positions, and dynamic control optimization of acupuncture positions is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of acupuncture efficacy recognition, and in particular to an attention neural network acupuncture control optimization system based on fNIRS guidance. Background Art

[0002] The statements in this section merely mention background art related to the present invention and do not necessarily constitute prior art.

[0003] With the development and progress of society, the improvement of people's material living standards and quality of life, the prevalence and disability rate of stroke have been increasing year by year. According to reports, in the analysis of global population mortality and disability rates, the incidence and disability rate of stroke ranked second and third in the world respectively. Acupuncture treatment, as an important part of TCM diagnosis and treatment, has been widely used in "cerebral stroke" in clinical practice and is currently an effective method for post-stroke rehabilitation treatment. "Getting Qi" is usually considered to be the key factor for the effectiveness of acupuncture. The so-called "getting Qi" refers to the needle sensation, which refers to the "sour, numb, swollen" and other feelings produced by the acupunctured person after acupuncture. On the other hand, during the acupuncture process, acupuncturists will also feel a heavy and tight feeling under the needle. However, "getting Qi" as a subjective feeling lacks objective representation. The energy, attention, technique of the acupuncturist and the quality of the acupuncture needle can all affect the feeling of "getting Qi". TCM theory believes that "Qi is the fullness of life" and "Qi" represents vitality and energy, which is one of the important factors constituting human life activities. Therefore, whether acupuncture can obtain qi affects the therapeutic effect of acupuncture on the patient's disease to a certain extent. Clinically, some stroke patients have speech and / or cognitive disorders and consciousness disorders. For such patients, they are unable to describe or accurately describe the feeling of acupuncture and give timely and effective feedback. Finding an effective and accurate method to identify "qi" is of great significance to both patients and clinical acupuncturists. Accurately identifying acupuncture "qi" can provide acupuncturists with timely and effective feedback, so as to adjust the acupuncture position, depth and technique in time. At the same time, patients can also obtain more effective rehabilitation treatment and shorten hospitalization time, which can reduce the economic burden of patients and avoid the waste of medical resources.

[0004] The brain center is the sensing site of acupuncture "De Qi". Studies have shown that acupuncture De Qi can activate or negatively activate different brain regions, but the results are still controversial, and the correlation between subjective De Qi and brain functional response has not yet been clarified. Functional near-infrared spectroscopy (fNIRS) is an emerging brain functional imaging technology with higher temporal resolution than functional magnetic resonance imaging, and has the advantages of safety, non-invasiveness, and portability. Its principle is to measure the hemodynamic response by measuring the change in hemoglobin concentration in the brain through near-infrared light, which indirectly reflects the brain's neural activity. In recent years, fNIRS technology has been applied in the field of acupuncture. In previous near-infrared studies, most of the changes in brain activation or brain functional connectivity after acupuncture treatment were explored by analyzing fNIRS data. A very small number of studies have found that acupuncture De Qi can cause activation of specific parts (positive activation of the frontal lobe). The frontal lobe is mainly related to mental, thinking and cognitive functions. The activation of the frontal lobe by De Qi indicates that mental, thinking and cognitive related areas may play an important role in the process of acupuncture De Qi. This may be related to the subjects' "keeping their minds" during acupuncture, that is, focusing their attention on feeling the needle sensation. However, there is no effective method to objectively evaluate whether acupuncture treatment is effective. Summary of the invention

[0005] In order to address the deficiencies of the prior art, the present invention provides an acupuncture dynamic control optimization system based on EEG signals and attention neural networks; computer technology is used to assist acupuncture workers in finding more accurate acupuncture positions, thereby achieving dynamic control optimization of acupuncture positions.

[0006] The fNIRS-guided attention neural network acupuncture control optimization system includes:

[0007] An acquisition module, which is configured to: acquire physiological signals of the brain of the subject during acupuncture;

[0008] A preprocessing module is configured to: preprocess the acquired physiological signals to obtain signal characteristics of each brain region;

[0009] A prediction module is configured to: input the signal features of each brain region into the trained attention neural network, and output a quantitative prediction result of acupuncture efficacy of each brain region;

[0010] The control module is configured to control the duration and position of acupuncture according to the quantitative prediction results of the acupuncture efficacy of each brain region.

[0011] The above technical solution has the following advantages or beneficial effects:

[0012] Through fNIRS, the brain area signal characteristics of the subjects in the "De Qi" state are obtained, and the accuracy of acupuncture "De Qi" recognition is improved based on the fNIRS-guided attention multimodal deep neural network, thereby solving the problems of inaccurate and non-objective acupuncture "De Qi" recognition; computer technology is used to assist acupuncture staff in finding more accurate acupuncture positions, thereby realizing dynamic control and optimization of acupuncture positions. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0014] Figure 1 Schematic diagram of the fNIRS installation location;

[0015] Figure 2 is a flow chart of the method of the present invention;

[0016] Among them, 1, the first channel, 2, the second channel, 3, the third channel, 4, the fourth channel, 5, the fifth channel, 6, the sixth channel, 7, the seventh channel, 8, the eighth channel, 9, the ninth channel, 10, the tenth channel, 11, the eleventh channel, 12, the twelfth channel, 13, the thirteenth channel, 14, the fourteenth channel, 15, the fifteenth channel, 16, the sixteenth channel, 17, the seventeenth channel, 18, the eighteenth channel, 19, the nineteenth channel, 20, the twentieth channel, 21, the twenty-first channel, 22, the twenty-second channel, 23, the twenty-third channel, 24, the twenty-fourth channel, 25, the twenty-fifth channel, 26, the twenty-sixth channel, 27, the twenty-seventh channel, 28, the twenty-eighth channel, 29, the twenty-ninth channel, 30, the thirtieth channel, 31, the thirty-first channel, 32, the thirty-second channel, 33, the thirty-third channel, 34, the thirty-fourth channel. DETAILED DESCRIPTION

[0017] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0018] Embodiment 1

[0019] like Figure 2 As shown, this embodiment provides an attention neural network acupuncture control optimization system based on fNIRS guidance, including:

[0020] An acquisition module, which is configured to: acquire physiological signals of the brain of the subject during acupuncture;

[0021] A preprocessing module is configured to: preprocess the acquired physiological signals to obtain signal characteristics of each brain region;

[0022] A prediction module is configured to: input the signal features of each brain region into the trained attention neural network, and output a quantitative prediction result of acupuncture efficacy of each brain region;

[0023] The control module is configured to control the duration and position of acupuncture according to the quantitative prediction results of the acupuncture efficacy of each brain region.

[0024] Furthermore, the step of obtaining physiological signals from the subject's brain during acupuncture includes:

[0025] Electrodes were attached to the epidermis of the subjects' heads, and a functional near-infrared spectroscopy imager was used to obtain physiological signals of the subjects' brains during acupuncture; the brains included: frontal lobe, temporal lobe, motor area, and occipital lobe.

[0026] It should be understood that Figure 1 As shown, the subjects wore a portable 34-channel near-infrared device. The device uses functional near-infrared spectroscopy (fNIRS) to obtain changes in oxygenated hemoglobin and deoxygenated hemoglobin during brain activity. The sampling frequency is 10 Hz, and the wavelengths are 760 nm and 850 nm. The portable 34-channel near-infrared device consists of 22 signal sources and 12 detectors. The first channel, the second channel, the third channel, the fourth channel, and the fifth channel are distributed in the left prefrontal cortex (LPFC), the sixth channel, the seventh channel, the eighth channel, the ninth channel, and the tenth channel are distributed in the right prefrontal cortex (RPFC), the eleventh channel, the twelfth channel, the thirteenth channel, the fourteenth channel, the fifteenth channel, the sixteenth channel, the seventeenth channel, the eighteenth channel, the nineteenth channel, and the twentieth channel are distributed in the left sensorimotor cortex (LSMC), and the twenty-first channel, the twenty-second channel, the twenty-third channel, the twenty-fourth channel, the twenty-fifth channel, the twenty-sixth channel, the twenty-seventh channel, the twenty-eighth channel, the twenty-ninth channel, and the thirty-fourth channel are distributed in the right sensorimotor cortex (RSMC). The 30th and 31st channels are distributed in the left occipital lobe (LOL), and the 32nd and 33rd channels are distributed in the right occipital lobe (ROL).

[0027] Furthermore, the acquired physiological signals are preprocessed to obtain signal characteristics of each brain region, including:

[0028] For physiological signals, time intervals and motion artifacts irrelevant to the trial were removed;

[0029] Filter physiological noise and interference signals of physiological signals;

[0030] converting the optical density value of the physiological signal into the concentration of oxygenated hemoglobin and the concentration of deoxygenated hemoglobin;

[0031] The oxygenated hemoglobin concentration time series and the deoxyhemoglobin concentration time series are input into the generalized linear model, and the activation coefficient of each brain region is output. The activation coefficient of each brain region is used as the signal feature of each brain region.

[0032] Furthermore, the filtering of the physiological noise and interference signal of the physiological signal is to filter the physiological noise and interference signal using Butterworth bandpass filtering, filter out the high-pass 0.01 Hz and low-pass 0.2 Hz signals, and retain the signals in the frequency range of 0.01-0.08 Hz.

[0033] Furthermore, the conversion of the optical density value of the physiological signal into the oxygenated hemoglobin concentration and the deoxygenated hemoglobin concentration is achieved by using a modified Beer-Lambert law formula;

[0034] The modified Beer-Lambert law formula:

[0035] (1)

[0036] (2)

[0037] (3)

[0038] in, is the change in light absorption, is the linear distance between the probes, is the wavelength-dependent extinction coefficient, DPF is the effective length from the light source to the detector, is the relative concentration change of HbO, = is the relative concentration change of HbR. The optical density value is converted into oxygenated hemoglobin HbO and deoxygenated hemoglobin HbR values.

[0039] Furthermore, the oxygenated hemoglobin concentration time series and the deoxyhemoglobin concentration time series are input into a generalized linear model, the activation coefficient of each brain region is output, and the activation coefficient of each brain region is used as the signal feature of each brain region; wherein,

[0040] Generalized Linear Models:

[0041] ;

[0042] Among them, y is the explained variable, , , …, is the characteristic variable, , ,…, The model coefficients corresponding to each explanatory variable, is the residual.

[0043] The HbO time series data were analyzed to obtain the activation coefficient β value of the whole brain (frontal lobe, temporal lobe, motor area, and occipital lobe).

[0044] Furthermore, the signal features of each brain region are input into the trained attention neural network, and the quantitative prediction results of acupuncture efficacy of each brain region are output; wherein the trained attention neural network includes:

[0045] N parallel branches; each branch is used to process the signal collected by a corresponding electrode;

[0046] Each branch includes: a multimodal learning network model DCNN and an attention module FGA connected in sequence;

[0047] The multimodal learning network model DCNN comprises: an input layer, a first convolutional layer, a first sampling layer, a second convolutional layer, a second sampling layer, a third convolutional layer, a third sampling layer, a first fully connected layer, a second fully connected layer and a third fully connected layer connected in sequence;

[0048] The attention module FGA includes: a fourth convolutional layer, a first global average pooling layer, a fifth convolutional layer, a second global average pooling layer, a fourth fully connected layer and a fifth fully connected layer which are connected in sequence.

[0049] Furthermore, the first convolution layer includes 6 convolution kernels, each of which has a size of 4×4, a stride of 1, and is padded with 0. The first sampling layer adopts a max-pooling operation, and the size of each window is 3×3 with a stride of 2. The second convolution layer includes 6 convolution kernels, each of which has a size of 5×5, a stride of 1, and is padded with 0. The second sampling layer adopts a max-pooling operation, and the size of each window is 2×2 with a stride of 2. The third convolution layer includes 8 convolution kernels, each of which has a size of 5×5, a stride of 1, and is padded with 0. The third sampling layer adopts a max-pooling operation, and the size of each window is 2×2 with a stride of 2. The number of neurons in the first fully connected layer is 256, the number of neurons in the second fully connected layer is 128, and the number of neurons in the third fully connected layer is 64.

[0050] Furthermore, the fourth convolutional layer includes 3 convolution kernels, each of which has a size of 3×3, a step size of 1, and is padded with 0. The first global average pooling layer adopts the global average pooling operation, and the size of each window is 3×3 and the step size is 2. The fifth convolutional layer includes 6 convolution kernels, each of which has a size of 3×3, a step size of 1, and is padded with 0. The second global average pooling layer adopts the global average pooling operation, and the size of each window is 3×3 and the step size is 2. The number of neurons in the fully connected layer F1 is 64, and the number of neurons in the fourth fully connected layer and the fifth fully connected layer is 32.

[0051] Furthermore, the multimodal learning network model DCNN is used to extract functional near-infrared spectral signal features of brain regions related to acupuncture, expand the feature range through three convolutional layers, remove redundant information through three sampling layers, and obtain the feature mapping space through three fully connected layers.

[0052] Furthermore, the attention module FGA is used to extract the attention features of acupuncture activation in the functional near-infrared spectroscopy signal at a deeper level, emphasizing the important areas of classification; deeper features are obtained through two convolutional layers, the receptive field is enhanced through two global average pooling layers, and the attention feature mapping space is obtained through two fully connected layers.

[0053] Furthermore, the signal features of each brain region are input into the trained attention neural network, and the quantitative prediction results of acupuncture efficacy of each brain region are output; wherein the training process of the trained attention neural network includes:

[0054] Constructing a training set, wherein the training set includes: a plurality of brain region signal features, each of which includes a known acupuncture efficacy quantitative label;

[0055] The training set is input into the attention neural network, and the attention neural network is trained. When the loss function value of the attention neural network no longer decreases, or the number of iterations exceeds the set number, the training is stopped to obtain the trained attention neural network.

[0056] Furthermore, the loss function for:

[0057] ;

[0058] in, is the total number of features, is the input fNIRS feature, For the prediction results.

[0059] It should be understood that the quantitative labels of acupuncture efficacy in the training set are manually annotated.

[0060] Furthermore, according to the quantitative prediction results of acupuncture efficacy in each brain region, the duration and location of acupuncture are controlled, including:

[0061] The quantitative labeling of acupuncture efficacy is divided into five levels, represented by numbers from 1 to 5;

[0062] Among them, the number 1 means no effect, the number 2 means slight effect, the number 3 means moderate effect, the number 4 means obvious effect, and the number 5 means very effective.

[0063] If the sum of the numbers corresponding to the acupuncture efficacy quantitative labels of all electrode positions is greater than or equal to the set threshold, it means that the current acupuncture position is effective, and an effective prompt is output to increase the acupuncture time;

[0064] If the sum of the numbers corresponding to the acupuncture efficacy quantification labels of all electrode positions is less than the set threshold, it means that the current acupuncture position is invalid, and an invalid prompt is output so that the acupuncture position can be changed.

[0065] The present invention uses fNIRS to obtain the brain regions that need to be paid attention to in the "De Qi" state of the subject, combines the brain region characteristics with the neural network, and establishes an attention neural model guided by fNIRS. By establishing a loss function, the detection model can obtain the optimal result of "De Qi" recognition according to the subject's response during the detection process, thereby improving the recognition accuracy of acupuncture "De Qi" and thus solving the problems of inaccurate and non-objective recognition of acupuncture "De Qi".

[0066] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An fNIRS-guided attention neural network acupuncture control optimization system, characterized by: include: An acquisition module is configured to: use a functional near-infrared spectroscopy imager to acquire physiological signals of various brain regions of the subject during acupuncture; Wherein, the brain regions include: frontal lobe region, temporal lobe region, motor region and occipital lobe region; A preprocessing module is configured to: preprocess the acquired physiological signals to obtain signal characteristics of each brain region; The preprocessing of the acquired physiological signals to obtain the signal characteristics of each brain region includes: For physiological signals, time intervals and motion artifacts irrelevant to the trial were removed; Filter physiological noise and interference signals of physiological signals; converting the optical density value of the physiological signal into the concentration of oxygenated hemoglobin and the concentration of deoxygenated hemoglobin; Input the oxygenated hemoglobin concentration time series and the deoxyhemoglobin concentration time series into the generalized linear model, output the activation coefficient of each brain region, and use the activation coefficient of each brain region as the signal feature of each brain region; in, The generalized linear model is: ； Among them, y is the explained variable, , , …, is the characteristic variable, , ,…, The model coefficients corresponding to each explanatory variable, is the residual; The filtering of the physiological noise and interference signal of the physiological signal is to filter the physiological noise and interference signal by applying Butterworth bandpass filtering, filtering out the high-pass 0.01 Hz and low-pass 0.2 Hz signals, and retaining the signals within the frequency range of 0.01-0.08 Hz; A prediction module is configured to: input the signal features of each brain region into the trained attention neural network, and output a quantitative prediction result of acupuncture efficacy of each brain region; Wherein, the trained attention neural network includes: N parallel branches; each branch is used to process the signal collected by a corresponding electrode; Each branch includes: a multimodal learning network model DCNN and an attention module FGA connected in sequence; The multimodal learning network model DCNN includes: an input layer, a first convolutional layer, a first sampling layer, a second convolutional layer, a second sampling layer, a third convolutional layer, a third sampling layer, a first fully connected layer, a second fully connected layer and a third fully connected layer connected in sequence; The multimodal learning network model DCNN is used to extract the functional near-infrared spectral signal features of the brain area related to acupuncture, expand the feature range through three convolution layers, remove redundant information through three sampling layers, and obtain the feature mapping space through three fully connected layers; The attention module FGA comprises: a fourth convolutional layer, a first global average pooling layer, a fifth convolutional layer, a second global average pooling layer, a fourth fully connected layer and a fifth fully connected layer connected in sequence; The attention module FGA is used to extract the attention features activated by acupuncture in the functional near-infrared spectroscopy signal at a deeper level, emphasizing the important areas of classification; obtains deeper features through two convolutional layers, enhances the receptive field through two global average pooling layers, and obtains the attention feature mapping space through two fully connected layers; The control module is configured to control the duration and position of acupuncture according to the quantitative prediction results of the acupuncture efficacy of each brain region.

2. The fNIRS-guided attention neural network acupuncture control optimization system as claimed in claim 1, characterized in that: The conversion of the optical density value of the physiological signal into the oxygenated hemoglobin concentration and the deoxygenated hemoglobin concentration is achieved by using a modified Beer-Lambert law formula; The modified Beer-Lambert law formula: (1) (2) (3) in, is the change in light absorption, is the linear distance between the probes, is the wavelength-dependent extinction coefficient, DPF is the effective length from the light source to the detector, is the relative concentration change of HbO, is the relative concentration change of HbR; the optical density value is converted into oxygenated hemoglobin HbO and deoxygenated hemoglobin HbR values.

3. The fNIRS-guided attention neural network acupuncture control optimization system as claimed in claim 1, characterized in that: The first convolutional layer includes 6 convolution kernels, each of which has a size of 4×4, a stride of 1, and is padded with 0. The first sampling layer adopts the maximum pooling operation, the size of each window is 3×3, and the stride is 2. The second convolutional layer includes 6 convolution kernels, each of which has a size of 5×5, a stride of 1, and is padded with 0. The second sampling layer adopts the maximum pooling operation, the size of each window is 2×2, and the stride is 2. The third convolutional layer includes 8 convolution kernels, each of which has a size of 5×5, a stride of 1, and is padded with 0. The third sampling layer adopts the maximum pooling operation, the size of each window is 2×2, and the stride is 2. The number of neurons in the first fully connected layer is 256, the number of neurons in the second fully connected layer is 128, and the number of neurons in the third fully connected layer is 64.

4. The fNIRS-guided attention neural network acupuncture control optimization system as claimed in claim 1, characterized in that: The fourth convolutional layer includes 3 convolution kernels, each of which has a size of 3×3, a stride of 1, and is padded with 0. The first global average pooling layer uses a global average pooling operation, and the size of each window is 3×3 with a stride of 2. The fifth convolutional layer includes 6 convolution kernels, each of which has a size of 3×3, a stride of 1, and is padded with 0. The second global average pooling layer uses a global average pooling operation, and the size of each window is 3×3 with a stride of 2. The number of neurons in the fully connected layer F1 is 64, and the number of neurons in the fourth and fifth fully connected layers is 32.

5. The fNIRS-guided attention neural network acupuncture control optimization system as claimed in claim 1, characterized in that: The signal characteristics of each brain region are input into the trained attention neural network, and the quantitative prediction results of acupuncture efficacy of each brain region are output; wherein the training process of the trained attention neural network includes: Constructing a training set, wherein the training set includes: a plurality of brain region signal features, each of which includes a known acupuncture efficacy quantitative label; Input the training set into the attention neural network, train the attention neural network, and stop training when the loss function value of the attention neural network no longer decreases or the number of iterations exceeds the set number to obtain the trained attention neural network; Loss Function for: ； in, is the total number of features, is the input fNIRS feature, For the prediction results.

6. The fNIRS-guided attention neural network acupuncture control optimization system as claimed in claim 1, characterized in that: According to the quantitative prediction results of acupuncture efficacy in each brain region, the duration and location of acupuncture are controlled, including: The quantitative labeling of acupuncture efficacy is divided into five levels, represented by numbers from 1 to 5; Among them, the number 1 means no effect, the number 2 means slight effect, the number 3 means moderate effect, the number 4 means obvious effect, and the number 5 means very effective. If the sum of the numbers corresponding to the acupuncture efficacy quantitative labels of all electrode positions is greater than or equal to the set threshold, it means that the current acupuncture position is effective, and an effective prompt is output to increase the acupuncture time; If the sum of the numbers corresponding to the acupuncture efficacy quantification labels of all electrode positions is less than the set threshold, it means that the current acupuncture position is invalid, and an invalid prompt is output so that the acupuncture position can be changed.

Images