CN114587389A

CN114587389A - Auxiliary diagnosis system suitable for patients with nervous dyskinesia and use method thereof

Info

Publication number: CN114587389A
Application number: CN202210197323.9A
Authority: CN
Inventors: 马辛; 高硕�; 卢玉姣
Original assignee: Beihang University
Current assignee: Beihang University
Priority date: 2022-03-02
Filing date: 2022-03-02
Publication date: 2022-06-07

Abstract

The invention discloses an auxiliary diagnosis system suitable for a rehabilitation stage of a patient with nervous movement disorder, which mainly comprises a hardware system with a myoelectric and muscle force signal synchronous acquisition function and an auxiliary diagnosis method of the patient with the nervous movement disorder based on a deep learning algorithm. Wherein the hardware system comprises: the device comprises a sensor module with an electromyographic signal acquisition function, a sensor module with muscle force signal detection and a signal processing module. The invention utilizes the muscle force signal to determine the contact state of the arm ring and the skin, pre-calibrates the electromyographic signal, matches and corrects the electromyographic signal by using the muscle force signal, and clearly defines the channel number of the electromyographic signal corresponding to the muscle group. The method comprises the steps of carrying out data analysis processing on signals acquired by a healthy side and an affected side of a patient synchronously, then carrying out difference feature extraction on the healthy side and the affected side, selecting features strongly associated with disease rehabilitation as a data set, inputting the data set into a neural network for model training, diagnosing the rehabilitation condition of a patient with the nervous dyskinesia, and building an assisted medical internet of things. The auxiliary diagnosis system has the characteristics of small volume, portability, simple operation, high accuracy and the like.

Description

Auxiliary diagnosis system suitable for patients with nervous dyskinesia and use method thereof

Technical Field

The invention relates to the field of rehabilitation diagnosis of electromyographic signals and nervous system diseases, in particular to a device for multi-channel electromyographic detection and flexible pressure detection, a related signal processing method and a patient auxiliary diagnosis method.

Background

When the motor nerve of the body is excited, nerve impulse is transmitted to the motor nerve ending to release neurotransmitter, the neurotransmitter and corresponding receptors on the muscle cell membrane change the permeability of potassium ion channels on the muscle cell membrane, the inner flow of sodium ions is far larger than the outer flow of potassium ions, action potential is generated, and the action potential is transmitted to the whole muscle cell membrane. When the muscle cell membrane is excited, the electric energy is converted into mechanical energy, so that myofibrils are shortened, the muscle is contracted, and limbs and bones are pulled to bend and move.

In recent years, many researchers at home and abroad have tried to develop studies on the pathogenesis and development stage of diseases with impaired neurological motor function by using myoelectric signals generated during human body exercise. The surface electromyography (sEMG) evaluation technology is mainly used for researching electrical signals of a neuromuscular system, has the characteristics of real-time objective and dynamic sensitivity, and can quantitatively reflect the muscle function change, the muscle strength level and the muscle cooperative contraction function to a great extent when the myoelectric signal activity changes under the condition of good control.

In the field of clinical assessment of neuromuscular diseases, Carlo Frigo et al, in 2009, used surface electromyography for disease classification, localization of focal lesions, detection of pathophysiological mechanisms and functional assessment, providing relevant information. In addition, in order to carry out mechanism research on nervous motor system diseases, domestic and foreign research institutions carry out mass acquisition and analysis on electromyographic signals of walking and arm appointed movement of a patient, and provide various muscle coordination models, for example, a Li Jian Hua professor team of Zhejiang university carries out statistical analysis on the activation sequence and strength of muscle groups in walking of the patient at a specific stage, and provides several typical muscle coordination action modes of a cerebral apoplexy patient; the subject group of the professor Mehrabi N of the university of Washington, USA and the subject group of the professor Asama H of the university of Tokyo, Japan are based on theories of motion module, optimization control and the like, and the system muscle groups of the patient in each motion stage are divided by using data obtained from a large number of experiments, so that a plurality of statistical models for evaluating the motion performance of the patient are established.

Currently, studies based on surface electromyographic signals have achieved a lot of research results, which have helped doctors and patients better understand the progress of the disease in relevant clinical applications. However, the further research and analysis of the motor rehabilitation state of the patient with motor dysfunction based on the myoelectric communication is seriously influenced by electromagnetic interference, motion artifacts, electrode offset, individual difference and the like, so that the reliability of the electromyographic signals is reduced to a certain extent, and an effective method for solving the problems of verification and correction of the electromyographic signals is not available at present.

Therefore, in the process of nervous system disease rehabilitation diagnosis and treatment based on electromyographic signal measurement, how to solve the problems of stability and reliability of the electromyographic signal is a key for further improving the accuracy of the acquired electromyographic signal and further providing more accurate and more dimensional information for human motion function analysis and patient disease rehabilitation condition analysis.

The contraction and the relaxation of the muscle bring the change of the muscle volume, and the pressure, the position, the size, the generation and the disappearance time of the pressure can be generated between the flexible sensors which are tightly attached to the flexible sensors, so that the final action effect of the nervous system on the human motion control can be reflected to a certain degree. And recording the time-space distribution of the pressure to obtain a surface pressure signal when the muscle moves. There is a certain internal correlation between the electromyographic signals and the pressure signals caused by the change of muscle volume, and the electromyographic signals are ahead of the muscle force. Therefore, the pressure signal generated between the contraction and relaxation of the muscle and the sensor can realize the correction of the electromyographic signal.

Compared with other movement-related physiological signals, the muscle strength signal generated between the muscle movement and the pressure sensor has good electromagnetic interference resistance, is portable, convenient to use and low in cost, is suitable for large-scale popularization, and can be widely applied to the fields of clinical diagnosis, man-machine interaction, artificial limb control, movement recognition, rehabilitation and medical treatment and the like.

Patients with nervous movement disorder (such as cerebral apoplexy) have asymmetric muscle contraction between the limbs at the healthy side and overcompensation and activation of muscle groups at the healthy side lower limbs due to the lack of sensory-motor information integration and the loss of posture control flexibility. The development degrees of the disease on the health-affected side of the patients in different development stages are different, and if myoelectric and muscle strength signals on the health-affected side are detected simultaneously, the pathogenesis and development state of the patients with the disease in the rehabilitation stage can be deeply researched.

Disclosure of Invention

The invention aims to provide an auxiliary diagnosis and treatment system suitable for a rehabilitation stage of a patient with the neural motor dysfunction, which mainly comprises a hardware system with a myoelectric and muscle force signal synchronous acquisition function and a neural network-based neural motor dysfunction patient rehabilitation diagnosis method. The equipment has the advantages of being fit with a human body curve, wearable, flexible and portable, low in cost and the like, and the method can be used for realizing auxiliary diagnosis of the rehabilitation condition of a nervous system disease such as a cerebral apoplexy patient.

In order to achieve the purpose, the invention provides the following scheme: firstly, the invention provides a flexible multi-channel pressure sensing array and a multi-channel electromyographic signal acquisition device based on a shunt mode (shunt mode).

Further, the pressure sensing array is composed of two layers (top layer and bottom layer) of high molecular material Thermoplastic polyurethane elastomer (TPU) sensing layers with better wear resistance, elasticity and flexibility and an adhesion layer (bonding the top layer and the bottom layer in a specific adhesion mode by an adhesive and generating a gap between the two layers) between the middle separation layer and the top layer.

Furthermore, the two TPU films of the pressure sensing array are printed with custom-made sensing electrodes (which may be circular, rectangular, etc.) in required number, and the sensing array is formed by a sensing pad formed by mixing conductive silver paste, pressure sensitive materials (such as FSR graphite), etc. Besides the sensing electrodes, the two layers of TPU films are respectively covered with a single-layer lead-out wire for connecting the sensing array and providing an interface for connecting the sensing array with an external circuit.

Furthermore, in order to reduce the number of outgoing lines, the pressure sensing arrays on the top layer are controlled in an array mode, are connected together in the same row and are led out through a conductive silver paste line; the pressure sensing arrays on the bottom layer are connected together in the same row and are also led out through a conductive silver paste line, and the corresponding electrode is selected by switching on the row and the row where the electrode is positioned, and the pressure value of the corresponding electrode is measured. In addition, consider that the muscle diameters of the limbs of the human body exhibit a non-constant distribution of thin-thick-thin in the skeletal direction. In order to reduce the influence of the prestress of the thicker muscle part on the pressure measurement result when the sensing film is fixed, the row-column spacing of different electrodes is customized according to the actual condition of a patient, so that the electrodes can better fit the curvature of four limbs of the human body, and the pressure acquisition result with higher signal-to-noise ratio is obtained.

Furthermore, an adhesive is placed at the edge of the two layers of pressure sensing arrays, and the two layers of pressure sensing arrays are bonded, so that the top layer and the bottom layer of conductive pads are not connected with each other under the condition that the sensing arrays are not subjected to any external force. When a single sensing point is subjected to external force from small to big, the surface area of the top layer in contact with the conducting pad of the bottom layer is gradually increased, so that the conductivity of the device is increased, and the resistance is reduced; when the method is applied to muscle force mode detection, when a pressure distribution mode exists, a plurality of sensing points are stressed by pressures with different magnitudes, and the resistance conditions of different sensing points are inconsistent.

The multichannel electromyographic signal acquisition equipment, namely a multichannel electromyographic signal front-end acquisition board, is composed of a plurality of electromyographic signal front-end acquisition board arrays and mainly comprises two parts, a metal electrode and a signal processing circuit.

Furthermore, the back of the front-end collecting plate is provided with three metal electrodes, one of the metal electrodes serves as a reference electrode, the other two metal electrodes form a differential electrode, each discrete front-end collecting plate can obtain one path of electromyographic signals, the front-end collecting plates are connected through an elastic rope, multi-channel electromyographic signal collection can be achieved, meanwhile, the elastic rope can be applied to various sizes, and the applicability of the system is improved.

Furthermore, a signal processing circuit is integrated on the front-end acquisition board and comprises a low-pass filter circuit for filtering motion artifacts, an instrument amplification circuit with a high common-mode rejection ratio, a band-pass filter for filtering high-frequency noise and power frequency interference and the like. Signals after primary processing by the front-end acquisition board are sent to a signal reading circuit, and sampling is carried out by an analog-to-digital converter (ADC) conversion chip, so that conversion from analog signals to digital signals is realized. Meanwhile, the output voltage of the resistance-voltage conversion circuit is collected by an analog-to-digital converter (ADC) controlled by a controller and is cached in a microcontroller; and under the control of the main control chip, the collected signals are sent to the upper computer in a Bluetooth transmission or serial port communication mode.

Besides the design of the device with the function of synchronously acquiring the multichannel pressure and the electromyographic signals, the invention also provides a rehabilitation stage diagnosis method for patients with motor function damage (caused by nervous system diseases) based on an independent component analysis method and a neural network. The model establishment steps are as follows:

the method comprises the following steps: a patient database is established. And recording diagnosis and treatment information of related diseases of the patient and a rehabilitation stage diagnosis result given by a doctor. And (4) defining a target muscle group, and selecting a fixed position for detection (different patients are adjusted according to a proportion). The type and the number of rehabilitation actions to be performed are determined according to the actual condition of the patient.

Step two: and measuring myoelectric signals and muscle pressure distribution patterns of the patient when the patient performs a specific rehabilitation action and uploading the myoelectric signals and muscle pressure distribution patterns to an upper computer. The power supply is turned on after the device is worn. Firstly, the arms are kept completely relaxed, after the output signal of the equipment is stable, the tightness degree of the equipment on the affected side is adjusted according to the output result of the muscle force signal, and the two sides are ensured to have the same pretightening force. The myoelectric signals and muscle strength signals of the patient during repeated rehabilitation actions are obtained by guiding the patient to repeat the designated rehabilitation actions (fist making, elbow bending, arm rotating, wrist lifting and the like) through voice. Firstly, the rehabilitation action 1 is measured, each action is repeated for 20 times (can be adjusted according to actual conditions), after the action is finished, the patient takes a rest for t seconds, the rehabilitation action 2 is measured, and each action is repeated for 20 times … … until all rehabilitation action measurements are finished.

Step three: calibrating the positions of the electromyographic signals of different paths according to the collected muscle force data, determining the channel numbers of the electromyographic signals of the corresponding muscles, and recording the channel data corresponding to the N paths of electromyographic signals of the healthy side as A₁＝{A₁₁,…,A_1NRecording data of channels corresponding to the healthy-side N-path muscle force signal sensors as B₁＝{B₁₁,…,B_1NRecording channel data corresponding to the N-path electromyographic signal sensor at the affected side as A₂＝{A₂₁,…,A_2NRecording channel data corresponding to the N-path muscle force signal sensor on the affected side as B₂＝{B₂₁,…,B_2N}. The myoelectricity sensor channel numbers of different test objects and the matching of muscle group positions are realized by utilizing a myoelectricity diagram generated by N paths of myoelectricity signal sensors, and the myoelectricity signal channels are reordered according to the myoelectricity signal characteristics and the like₁＝{C₁₁,…,C_1N}、C₂＝{C₂₁,…,C_2NAnd ensuring the matching of the myoelectric signal and the myoelectric signal positions during the test of different patients.

Step four: after the establishment process of the patient data set is completed, the data processing process is carried out, and the electromyographic signals are subjected to band-pass filtering at 10-500Hz and then subjected to 50Hz notch processing. And then, further denoising the electromyographic signals by adopting an Independent Component Analysis (ICA), firstly, performing principal component decomposition (PCA) on the m-path signal observation signals X, selecting the first n principal components according to the contribution rate, thus reducing the dimension of the signals, and performing PCA whitening. Whitening matrix is V_n×mThe whitened signal is Z ═ VX, Z has no correlation between dimensions and has a variance of 1. And setting the number of the independent components as n, and carrying out ICA decomposition on the whitening signal Z to obtain n mutually independent source components, wherein the n-dimensional separation signal is S-WZ, W is a separation matrix, and A is a mixing matrix. The useful signal is typically a super gaussian signal with a kurtosis value greater than zero, while the noise is typically a sub-gaussian signal with a kurtosis value less than 0. Therefore, the noise can be identified and distinguished by utilizing the positive and negative of the kurtosis value, so that the noise component is automatically identified by utilizing the positive and negative of the kurtosis threshold, the kurtosis is the fourth-order cumulant of a random variable, and the formula is kurt (y) E [ y [ -n-y [ -n⁴]-3(E[y²])². Reconstructing the processed dimensional signal S through a mixing matrix A and a whitening matrix V, restoring each component to a space signal, and reconstructing an m-dimensional signal Y-V^TAS。

Step five: the total number of the electromyographic signal sample points of each channel is recorded as n, the ith sample point data is recorded as D(i) And the sample frequency is denoted as f_sThe total duration of a sampling process is denoted T, i.e. n ═ f_sT. Fourier transforming the surface myoelectric signal can obtain the spectral distribution of the signal. Let the ith frequency component be p_iFor a total of M components. Extracting electromyographic signals { C of each channel according to the following formula₁₁,…,C_1N}、{C₂₁,…,C_2NAnd (4) corresponding characteristics of a time domain (maximum amplitude, minimum amplitude, amplitude range, root mean square), a frequency domain (peak frequency, average frequency, total power, median frequency) and the like.

Step six: and selecting signal characteristics strongly associated with the stroke patient. Aiming at multidimensional and huge electromyographic signal characteristic data, an association relation between the characteristic and a patient rehabilitation stage is mined and extracted by adopting an association rule mining algorithm, a valuable association rule is discovered by traversing and collecting the frequency of various characteristic combinations, and then M-dimension characteristic data strongly associated with the patient rehabilitation condition is determined to obtain a data set D. Recording the strong association M-dimensional characteristic data set of the healthy side and the affected side as D₁＝{D₁₁,…,D_1M}、D₂＝{D₂₁,…,D_2M}。

Step seven: using a method based on T statistic to respectively align healthy and affected side feature sets D₁、D₂Calculating to obtain the characteristic D_ij(

i

1, 2;

j

1,2 … M) corresponding to the statistic x^D1,y^D2. If feature D_ijThe expression is consistent on two data sets, and ideally, x should be used^D1＝y^D2I.e. point (x)^D1,y^D2) And the feature statistics which are consistent in expression are drawn on a two-dimensional plane on the basis of y, x and a two-dimensional rectangular coordinate system, namely, the points which are closest to y, x, and the relatively larger features are far points. The distance x may be used for any point in the two-dimensional plane

Indicating that a minimum distance value min _ distance is set when the minimum distance value min _ distance is satisfied

It indicates that the feature is sufficiently different on the two datasets, i.e., the robust and diseased signal features are different. Given that the feature points near the origin are closer to the centerline, the selection will be

The correction is | x-y |/| x + y |, so that the feature distribution is homogenized. And (3) judging by using | x-y |/| x + y | > or more than min _ distance, and reserving the features larger than the minimum distance so as to obtain a data set D'.

Step eight: the feature set D' strongly associated with the patient rehabilitation condition is used as training data, the diagnosis of the patient rehabilitation condition by a doctor is used as a label, the label is input into a Fully Connected Neural Network (Fully Connected Neural Network), a Neural Network model is obtained through training, and model parameters are stored. This process only needs to go on once in equipment use earlier stage, and the model after the training is preserved at the host computer, need not to carry out the model training again in the user use of later stage.

Step nine: the prediction process is repeatedly carried out in the using process, the acquired signals are amplified and preliminarily filtered, then transmitted to an upper computer through a signal transmission system, and input into a trained model after being processed by data, and the classification result of the rehabilitation stage is obtained.

Compared with the prior art, the invention has the following advantages:

1. the system disclosed by the invention has the function of synchronously acquiring the myoelectric signals and the muscle force signals, and provides accurate and high-resolution data support for correction of the myoelectric signals and rehabilitation detection and mechanism research of patients with the nerve motor function.

2. The device measures pressure signals caused by volume changes caused by muscle expansion, and can realize the correction of electromyogram by using a pressure diagram, thereby improving the accuracy and reliability of the utilization of the multichannel electromyogram signals.

3. The deep research on the rehabilitation condition and the pathogenesis is realized by researching the difference of the healthy and affected sides of the patients with the nervous dyskinesia.

4. The invention realizes the automatic auxiliary diagnosis of the rehabilitation condition of the patient with the nervous motor dysfunction.

5. The invention belongs to non-invasive measurement, does not cause wound, can be used for a long time and does not influence the daily life of a patient.

Drawings

Fig. 1 shows a schematic diagram of a laminated structure and connection mode of a flexible muscle force sensing front end.

Fig. 2 shows a structural design plan development view of the flexible multi-channel muscle force sensor.

Fig. 3a is a schematic perspective view of a front-end circuit collecting board with an electromyographic signal collecting function.

Fig. 3b is a schematic plan view of a front-end circuit collecting board with electromyographic signal collecting function.

Fig. 4 is a schematic plan view showing the whole hardware system of the present invention.

FIG. 5 is a flow chart illustrating a method of use of the present invention, including a data collection process, a data processing process, and a predictive classification process.

Description of the reference numerals

TPU layer 11: upper TPU layer

12: lower TPU layer 2: piezoresistive unit

21: pressure-sensitive material 22: spacer layer

23: metallic silver interdigital electrode 3: electrode wire

4: myoelectric electrode unit 41: metal electrode 1

42: reference electrode 43: metal electrode 2

5: a PCB board 6: connecting hole

7: connecting wire

S101-S109: using steps one to nine of the system

Detailed Description

The technical solutions in the embodiments of the present invention will be described in detail below with reference to the accompanying drawings in the embodiments of the present invention. It should be noted that this summary should not be construed as limiting the invention, and that the described embodiments are merely a subset of the embodiments of the invention, rather than a complete subset. The embodiments are only intended to help the understanding of the claims and the core idea of the present invention. For those skilled in the art, the invention is not limited to the specific embodiments and applications, but rather, may be modified within the scope of the invention as defined by the appended claims.

In the exemplary embodiment of the present invention, as shown in fig. 1 and fig. 2, the designed multichannel flexible muscle force sensor comprises an upper TPU layer 1, a lower TPU layer 1, a piezoresistive unit 2 and an electrode wire 3. Two interdigitated finger-inserting type electrode wires are printed on a substrate of the flexible muscle strength sensor, and an adhesive insulating layer is arranged at the position where the flexible muscle strength sensor is connected with the edge of the substrate and is used as a conductive wire protection area, so that a conductive wire is not easy to break; and the other substrate is covered with a sensing pad made of conductive silver paste and a pressure sensitive material. The top layer and the bottom layer are separated by a spacing layer. When the applied pressure rises, the gap between the interdigitated conductive lines is filled with a conductive pressure sensitive material, causing the conductance between the two conductive lines to rise. The piezoresistive sensor has high sensitivity and a large linear range to weak pressure, and is flexible and telescopic. Considering that the amplitude of the muscle force signal is small and the requirement on the detection precision is high, the piezoresistive sensor in the direct mode is selected as the muscle force sensing unit.

The pressure sensing array is composed of two high polymer material thermoplastic polyurethane elastomer rubber sensing layers with good wear resistance, elasticity and flexibility and an adhesion layer 1 in the middle of a middle separation layer. And respectively printing rectangular sensing electrodes on the two layers of TPU films to form an N x M type sensing array. Besides the sensing electrodes, the two layers of TPU films are respectively covered with a single-layer lead-out wire for connecting the sensing array and providing an interface for connecting the sensing array with an external circuit. According to the shape design, on one hand, the pressure sensing electrode is more convenient to control, and meanwhile, the pressure sensing electrode and the front-end acquisition board of the electromyographic sensor are distributed in a cross mode, so that the synchronous acquisition of the electromyographic signal and the pressure signal is realized, and the correlation of the two acquired signals is convenient to analyze, as shown in fig. 4.

The multichannel muscle force sensor has different pressure distribution modes when a patient does rehabilitation exercise, a plurality of sensing points can be stressed by pressure with different sizes, and the resistance conditions of different sensing points are inconsistent. And the two layers of muscle strength sensing arrays are connected to a low-power-consumption muscle strength array reading circuit through an FPC (flexible printed circuit) interface to acquire and cache data, and are uploaded to an upper computer to be visualized and further analyzed.

In order to reduce the influence of the prestress of the thicker muscle part on the pressure measurement result when the sensing film is fixed, the row-column spacing of different electrodes can be customized according to the actual condition of a patient so as to better fit the curvature of the limb of the human body and obtain a pressure acquisition result with higher signal-to-noise ratio. The array structure design of the curve fitting human body limb can fully record the pressure signal space distribution mode of the muscle at the healthy and affected side during the activity, such as the activity area, the signal size, the time sequence and other related information, and detect the coordination and the time sequence among different muscles by respectively wearing the structural form of the acquisition system at the healthy and affected side.

The multipath electromyographic signal acquisition system is mainly formed by connecting a plurality of front-end acquisition boards at equal intervals, as shown in fig. 3a and 3b, for each independent electromyographic signal front-end acquisition board, three metal electrodes 4 are arranged on the back surface, wherein an electrode 42 is used as a reference electrode, an electrode 41 and an electrode 43 form a differential electrode, each discrete front-end acquisition board can obtain one path of electromyographic signal, four corners of each front-end acquisition board are respectively provided with a circular hole 6 with the diameter of 1.5mm, the front-end acquisition boards are connected by an elastic rope 7, multi-channel electromyographic signal acquisition can be realized, the detection diameter range is adjustable, the system is suitable for various sizes, and the applicability of the system is improved.

As shown in fig. 3a, a signal conditioning circuit 5 is integrated on the back of the front-end acquisition board, and comprises a 10Hz high-pass filter circuit and a 500Hz low-pass filter circuit for filtering motion artifacts, an instrument amplification circuit with a high common-mode rejection ratio, a wave trap for filtering 50Hz power frequency interference, and the like. Signals after the front end acquisition board preliminary treatment and the signals of multichannel pressure sensor output send into signal reading circuit, through the analog-to-digital conversion chip sampling, convert digital signal into, under main control chip's control, send into the host computer through bluetooth transmission or serial ports communication mode with the signal of gathering.

The specific steps of the method are described in detail below with reference to the flow chart.

S101: a patient database is established. And recording diagnosis and treatment information of related diseases of the patient and a rehabilitation stage diagnosis result given by a doctor. And (4) defining a target muscle group, and selecting a fixed position for detection (different patients are adjusted according to a proportion). The type and the number of rehabilitation actions to be performed are determined according to the actual condition of the patient.

S102: and measuring myoelectric signals and muscle pressure distribution patterns of the patient when the patient performs a specific rehabilitation action, and uploading the myoelectric signals and muscle pressure distribution patterns to an upper computer. The power supply is turned on after the equipment is worn. Firstly, the arms are kept completely relaxed, after the output signal of the equipment is stable, the tightness degree of the equipment on the affected side is adjusted according to the output result of the muscle force signal, and the two sides are ensured to have the same pretightening force. Then, the patient is guided to repeat the appointed rehabilitation actions (fist making, elbow bending, arm rotating, wrist lifting and the like) through voice, and the myoelectric signals and muscle strength signals of the patient are obtained when the patient repeatedly does the rehabilitation actions. The rehabilitation action 1 is measured, each action is repeated for 20 times (can be adjusted according to actual conditions), after the action is finished, the rehabilitation action 2 is measured after the rest of t seconds, and each action is repeated for 20 times … … until all rehabilitation action measurements are finished.

S103: calibrating the positions of the electromyographic signals of different paths according to the collected muscle force data, determining the channel numbers of the electromyographic signals of the corresponding muscles, and recording the channel data corresponding to the N paths of electromyographic signals of the healthy side as A₁＝{A₁₁,…,A_1NRecording data of channels corresponding to the healthy-side N-path muscle force signal sensors as B₁＝{B₁₁,…,B_1NRecording channel data corresponding to the N-path electromyographic signal sensor at the affected side as A₂＝{A₂₁,…,A_2NRecording channel data corresponding to the N-path muscle force signal sensor on the affected side as B₂＝{B₂₁,…,B_2N}。

S104: the myoelectricity sensor channel numbers and muscle group positions of different test objects are matched by using a myoelectricity diagram generated by N paths of myoelectricity signal sensors, and the myoelectricity signal channels are reordered according to the myoelectricity signal characteristics and the like₁＝{C₁₁,…,C_1N}、C₂＝{C₂₁,…,C_2NAnd ensuring the matching of the myoelectric signal and the myoelectric signal positions during the test of different patients.

S105: after the establishment process of the patient data set is completed, a data processing process is carried out, electromyographic signals are subjected to band-pass filtering at 10-500Hz, 50Hz notch processing is carried out, effective components of the signals are selected by a blind source separation method such as an independent component analysis method, and noise is further removed.

S106: the total number of electromyographic signal sample points of each channel is recorded as n, the ith sample point data is recorded as D (i), and the sample frequency is recorded as f_sThe total duration of a sampling process is denoted as T, i.e. n ═ f_sT. Fourier transforming the surface myoelectric signal can obtain the spectral distribution of the signal. Let i-th frequency component be p_iAnd N components. And respectively extracting characteristics of time domains (maximum amplitude, minimum amplitude, amplitude range, root mean square), frequency domains (peak frequency, average frequency, total power, median frequency) and the like corresponding to the electromyographic signals of the channels according to the following formula. Recording M-dimensional feature data sets corresponding to the healthy side and the affected side as D₁＝{MaxA₁₁…MaxA_1M,MinA₁₁…MinA_1M,RMS₁₁…RMS_1M,PF₁₁…PF_1M，MF₁₁…MF_1M，TP₁…TP_1M}，D₂＝{MaxA₂₁…MaxA_2M，MinA₂₁…MinA_2M，RMS₂₁…RMS_2M，PF₂₁…PF_2M，MF₂₁…MF_2M，TP2₁…TP_2M}，D₁And D₂Together constituting a data set D of the patient.

(1) The Maximum Amplitude (Max a) reflects the sample point with the Maximum signal Amplitude in the sample signal, and the calculation formula is as follows: MaxA max { D (1), D (2),.., D (n) }

The Minimum Amplitude (Min Amplitude, Min a) is reflected as the sample point with the Minimum signal Amplitude in the sample signal, and the calculation formula is as follows: min { D (1), D (2),. }, D (n) }

(2) Root Mean Square (RMS), which reflects the average energy of the electromyographic signals to some extent, is calculated as follows:

(3) peak Frequency (PF) represents the maximum power of the electromyographic signal, and the calculation formula is as follows:

PF＝max(p₁，p₂，…，p_N)

(4) the average Frequency (Mean Frequency, MF) represents the average Frequency of the power spectrum of the signal, and is calculated as follows:

(5) total Power (TP) represents the sum of the muscle electric signal Power spectral densities, and is calculated as follows:

s107: and (4) extracting difference characteristics of the diseased side. Using a method based on T statistic to respectively align healthy and affected side feature sets D₁、D₂Calculating to obtain the characteristic D_ij(

i

1, 2;

j

1,2 … M) corresponding to the statistic x^D1，y^D2. Calculating point (x)^D1，y^D2) And setting a minimum distance value min _ distance in a two-dimensional rectangular coordinate system when the distance y is equal to the distance x, and when the condition that | x-y |/| x + y | ≧ min _ distance is met, indicating that the difference of the characteristic on the two data sets is large enough, namely the difference of the signal characteristics of the healthy side and the affected side is large, and retaining the characteristic to obtain a characteristic set D' strongly associated with the rehabilitation condition of the patient.

S108: before signals are input into a neural network, data and tags are randomly scrambled, then dimension increasing is carried out on the data, the tags are converted into one-hot codes (one-hot), and then the data and the test sets are randomly divided into data sets and test sets according to the ratio of 4: 1. And (3) taking the extracted electromyographic signal characteristics as training data, taking the diagnosis of the patient rehabilitation condition by a doctor as a label, inputting the label into a neural network, training to obtain a neural network model, and storing model parameters. This process only needs to go on once in equipment use earlier stage, and the model after the training is preserved at the host computer, need not to carry out the model training again in the user use of later stage.

S109: the prediction process is repeatedly carried out in the using process, the acquired signals are amplified and primarily filtered, then transmitted to an upper computer through a signal transmission system, and input into a trained model after the data processing process, and a rehabilitation stage classification result is obtained.

The implementation mode of the technical scheme of the invention has the advantages of strong portability, convenient measurement process, non-invasive measurement and long-term use, and the diagnosis of the rehabilitation condition of the patient with the nervous dyskinesia is more convenient and quicker. In addition, the myoelectric signal calibration process enables the measurement to be more accurate and reliable. Other benefits and advantages of this embodiment will be apparent to those skilled in the art having the benefit of the teachings herein.

Representative embodiments of the present invention are described above in detail. It will be appreciated that modifications and variations of the inventive concept, such as positioning the device in different parts of the body for health monitoring, processing signals using other blind source separation methods or changing different neural network models, etc., may be made by those skilled in the art without the need for inventive faculty. Therefore, the technical solutions available to those skilled in the art through logic analysis, reasoning and limited experiments based on the prior art according to the concept of the present invention should be within the scope of protection defined by the claims.

Claims

1. An auxiliary diagnosis system suitable for rehabilitation stages of patients with nervous movement disorders mainly comprises a hardware system with myoelectric and muscle force signal synchronous acquisition functions and an auxiliary diagnosis method for rehabilitation conditions of patients with nervous movement disorders based on a neural network.

2. An auxiliary diagnostic system suitable for a patient with a neuromotor disorder, comprising:

the sensor module with the electromyographic signal acquisition function is a multi-channel electromyographic signal acquisition device, each channel comprises three metal electrodes which are uniformly distributed, two of the metal electrodes are detection electrodes, one of the metal electrodes is a reference electrode, the reference electrode is positioned between the two detection electrodes, and the two detection electrodes are connected with the signal processing module;

the sensor module is a flexible multi-channel pressure sensing array with special shape design and comprises array pressure signal acquisition electrodes designed according to a human body curve and leads made of conductive silver paste, and each electrode is connected with the power module and the signal processing module through the leads;

the signal processing module is mainly used for carrying out corresponding filtering processing on the electromyographic signals collected by the electromyographic signal sensor module, carrying out differential amplification on the signals, and meanwhile processing the signals from the myodynamia sensor module into voltage signals through the voltage division circuit. The signal processing module carries out digital-to-analog conversion on the myoelectric signals and the muscle strength signals and then transmits the signals to a computer through Bluetooth or a serial port for subsequent analysis and processing.

3. A neural network-based neural dyskinesia patient auxiliary diagnosis method is characterized in that electromyographic signals and muscle strength signals collected by a hardware system are processed, and then are matched with each other by using the muscle strength signals, so that the muscle strength signal channel collected each time is matched with a corresponding muscle group, and the muscle strength signals are used for correcting the electromyographic signals. The acquired signals are used as a training set after being processed by data and extracted by characteristics, the diagnosis result of a doctor on a patient is used as a label, the label is input into a neural network to train a network model, the mapping relation between the input signals and the rehabilitation condition of the patient is established, and when new signals are input, the predicted diagnosis result can be automatically output.

4. The flexible multichannel muscle force signal acquisition sensor module according to claim 2, wherein the sensor module comprises two layers (top layer and bottom layer) of high molecular material Thermoplastic polyurethane elastomer rubber (TPU) sensing layer with better wear resistance, elasticity and flexibility and an adhesive layer between the middle separation layers (the top layer and the bottom layer are bonded by an adhesive in a specific adhesion manner, and a gap is formed between the two layers). The two layers of TPU films are printed with sensing electrodes (which can be round, rectangular and the like) in a customized form and in required number, and the sensing array is formed by a sensing pad formed by mixing conductive silver paste, pressure sensitive materials (such as FSR graphite) and the like.

5. The sensor module for collecting the flexible multi-channel muscle strength signals according to claim 2, wherein the pressure sensing array on the top layer is controlled in an array manner, is connected together in the same row and is led out through a conductive silver paste line; the pressure sensing arrays on the bottom layer are connected together in the same row and are also led out through a conductive silver paste line. And (4) selecting the corresponding electrode by switching on the row and the column where the electrode is positioned, and measuring the pressure value of the corresponding electrode.

6. The sensor module for flexible multi-channel muscle strength signal acquisition as recited in claim 2, wherein the row-column spacing of the different electrodes is customized to fit the curvature of the extremities of the human body to reduce the effect of the pre-stress of the thicker muscle portion on the pressure measurement when the sensing membrane is secured.

7. The flexible multi-channel muscle force signal acquisition sensor module of claim 2, wherein the two layers of muscle force sensing arrays are connected by an adhesive placed at the edge. So that the top layer is not connected to the bottom layer conductive pads without any external forces on the sense array. When a single sensing point is subjected to external force from small to large, the surface area of the contact between the top layer and the bottom layer of the conducting pad is gradually increased, so that the conductivity between the conducting pads is increased, and the resistance is reduced.

8. The multi-channel electromyographic signal acquisition device of claim 2, consisting essentially of an electromyographic signal front-end acquisition board and a connection device, wherein the front-end acquisition board consists of a metal electrode mounted on the back and a signal processing circuit on the front, and essentially comprises a signal amplification circuit and a band-pass filter circuit. The connecting means (either a connecting wire or a connecting band may be used) connects the individual collection plates in series.

9. The method for aided diagnosis of a patient with a neuromotor disorder of claim 3, comprising the steps of:

the method comprises the following steps: a patient database is established. And recording diagnosis and treatment information of related diseases of the patient and a rehabilitation stage diagnosis result given by a doctor. And (4) defining a target muscle group, and selecting a fixed position for detection (different patients are adjusted according to a proportion). Determining the type and the number of rehabilitation actions to be executed according to the actual condition of a patient;

step two: and measuring myoelectric signals and muscle pressure distribution patterns of the patient when the patient performs a specific rehabilitation action, and uploading the myoelectric signals and muscle pressure distribution patterns to an upper computer. The power supply is turned on after the equipment is worn. Firstly, the arms are kept completely relaxed, after the output signal of the equipment is stable, the tightness degree of the equipment on the affected side is adjusted according to the output result of the muscle force signal, and the two sides are ensured to have the same pretightening force. Then, the patient is guided to repeat the appointed rehabilitation actions (fist making, elbow bending, arm rotating, wrist lifting and the like) through voice, and the myoelectric signals and muscle strength signals of the patient are obtained when the patient repeatedly does the rehabilitation actions. Measuring rehabilitation action 1, repeating the action for 20 times (which can be adjusted according to actual conditions) each time, after the action is finished, resting for t seconds, measuring rehabilitation action 2, and repeating the action for 20 times each time … … until all rehabilitation action measurements are finished;

step three: calibrating the positions of the electromyographic signals of different paths according to the collected muscle force data, determining the channel numbers of the electromyographic signals of the corresponding muscles, and recording the channel data corresponding to the N paths of electromyographic signals of the healthy side as A₁＝{A₁₁,…,A_1NRecording data of channels corresponding to the healthy-side N-path muscle force signal sensors as B₁＝{B₁₁,…,B_1NRecording channel data corresponding to the N-path electromyographic signal sensor at the affected side as A₂＝{A₂₁,…,A_2NRecording channel data corresponding to the N-path muscle force signal sensor on the affected side as B₂＝{B₂₁,…,B_2N}。The myoelectricity sensor channel numbers of different test objects and the matching of muscle group positions are realized by utilizing a myoelectricity diagram generated by N paths of myoelectricity signal sensors, and the myoelectricity signal channels are reordered according to the myoelectricity signal characteristics and the like₁＝{C₁₁,…,C_1N}、C₂＝{C₂₁,…,C_2NEnsuring the matching of the positions of the myoelectric signal and the myoelectric signal when different patients are tested;

step four: after the establishment process of the patient data set is completed, a data processing process is carried out, electromyographic signals are subjected to band-pass filtering at 10-500Hz, 50Hz notch processing is carried out, effective components of the signals are selected by a blind source separation method such as an independent component analysis method, and noise is further removed;

step five: the total number of electromyographic signal sample points of each channel is recorded as n, the ith sample point data is recorded as D (i), and the sample frequency is recorded as f_sThe total duration of a sampling process is denoted T, i.e. n ═ f_sT. Fourier transforming the surface myoelectric signal can obtain the spectral distribution of the signal. Let the ith frequency component be p_iFor a total of M components. Extracting electromyographic signals { C of each channel according to the following formula₁₁,…,C_1N}、{C₂₁,…,C_2NCorresponding characteristics of a time domain (maximum amplitude, minimum amplitude, amplitude range, root mean square), a frequency domain (peak frequency, average frequency, total power, median frequency) and the like;

step six: and selecting signal characteristics strongly associated with the stroke patient. Aiming at multidimensional and huge electromyographic signal characteristic data, an association relation between the characteristic and a patient rehabilitation stage is mined and extracted by adopting an association rule mining algorithm, a valuable association rule is discovered by traversing and collecting the frequency of various characteristic combinations, and then M-dimension characteristic data strongly associated with the patient rehabilitation condition is determined to obtain a data set D. Recording the strong association M-dimensional characteristic data set of the healthy side and the affected side as D₁＝{D₁₁,…,D_1M}、D₂＝{D₂₁,…,D_2M}；

Step seven: using a method based on T statistic to respectively align healthy and affected side feature sets D₁、D₂Calculating to obtain the characteristic D_ij(i 1, 2; j 1,2 … M) corresponding to the statistic x^D1,y^D2. Calculating point (x)^D1,y^D2) Setting a minimum distance value min _ distance in a two-dimensional rectangular coordinate system from a straight line y to x, and when the condition that | x-y |/| x + y | > is more than or equal to min _ distance is met, indicating that the difference of the characteristic on two data sets is large enough, namely the difference of the signal characteristics of the healthy side and the affected side is large, and keeping the characteristic;

step eight: and (3) taking the feature set D strongly associated with the rehabilitation condition of the patient as training data, taking the diagnosis of the patient rehabilitation condition by the doctor as a label, inputting the label into a neural network, training to obtain a neural network model, and storing model parameters. The process is only needed to be carried out once in the early stage of equipment use, the trained model is stored in an upper computer, and model training is not needed in the later-stage user use process;