Pelvic floor nerve muscle group image data analysis method based on deep learning

Abstract

The invention discloses a pelvic floor nerve muscle group image data analysis method based on deep learning, and relates to the technical field of medical instrument usage, six pelvic floor function modes are adopted, 20 features are adopted, resting, fast and slow movement and reflex activities are all included, and completeness of input information is ensured through 32 or 64 channels for example; meanwhile, in order to deal with the technical problem of dimension explosion, a deep learning method is adopted to reduce dimensions and extract main features, and relevance between the features is also enhanced, so that the situation that only a single-dimension pelvic floor function is carried out on a single feature in an existing method is avoided, and more pelvic floor diseases can be accurately analyzed; besides, high-density vagina electrodes with electrodes distributed in a uniform array mode continue to be used, actual operation and use are facilitated, analysis and positioning difficulty is transferred to algorithm steps, and practicability and operability of the whole system are improved.

Description

Technical Field

[0001] The present invention relates to the field of medical device use technology, and specifically to a pelvic floor nerve muscle group image data analysis method based on deep learning. Background Art

[0002] The pelvic floor muscles are composed of complex multiple muscle groups, mainly including the vaginal sphincter, urethral sphincter, external anal sphincter, puborectalis muscle, pubococcygeus muscle, iliococcygeus muscle, etc. These muscles are distributed in different directions and depths of the pelvic floor, supporting the three horizontal structures of the pelvic floor; in addition, the causes of pelvic floor diseases include neurological dysfunction, muscular dysfunction or both.

[0003] Traditional pelvic floor neuromuscular detection methods often use single-channel large-area electrode detection to obtain a single-dimensional electromyographic signal. This method can only detect the overall electrical information parameters of the entire pelvic floor muscle group, cannot subdivide the neuromuscular system in all directions and depths of the pelvic floor, and cannot focus on specific lesion sites. In addition, according to traditional pelvic floor neuromuscular detection methods, doctors can only rely on a single-dimensional diagnostic report and combine it with the patient's main symptoms for diagnosis. This method is highly dependent on the doctor's experience, which increases the probability of missed diagnosis or misdiagnosis. Summary of the invention

[0004] The purpose of the present invention is to provide a pelvic floor nerve muscle group image data analysis method based on deep learning to solve the problems raised in the prior art.

[0005] To achieve the above object, the present invention provides the following technical solution, a pelvic floor nerve muscle group image data analysis method based on deep learning, the analysis method comprising:

[0006] Step S1: using a vaginal pelvic floor probe to collect pelvic floor electromyography images;

[0007] Step S2: Collect sample points from the pelvic floor electromyography image and perform

[0008] Feature extraction;

[0009] Step S3: reduce the dimension of features and integrate the features;

[0010] Step S4: using the existing pelvic floor feature database, matching the reduced dimension features with the features in the database;

[0011] Step S5: Diagnose pelvic floor electromyography diseases by calculating the similarity between the features after dimensionality reduction and the samples.

[0012] Furthermore, the vaginal pelvic floor probe is attached with a channel for detecting abdominal muscle electromyography, and the electrode points are evenly distributed on the vaginal pelvic floor probe.

[0013] Furthermore, step S2 is to collect sample points from the pelvic floor electromyography image with a window of 50 ms and a step length of 10 ms, and calculate the root mean square of the electromyography signal in each channel of the original pelvic floor electromyography and the abdominal muscle channel according to the sample points. The calculation formula is as follows:

[0014]

[0015] Among them, i represents the number of channels, k represents the sampling point, and x i (k) represents the kth sampling point of the original electromyographic signal of the i-th channel, N represents the total sampling points in the window, RMS i Represents the RMS value of the electromyographic signal in channel i.

[0016] Furthermore, the pelvic floor electromyography image includes six stages, which are: a pre-resting stage, a fast contraction and release stage, a contraction stage, a 1-minute holding stage, a coughing stage, and a post-resting stage, and features of the six stages are extracted based on the root mean square.

[0017] Characteristics of the pre-resting phase include pre-resting potential and pre-resting variability;

[0018] The formula for calculating the front resting potential is as follows:

[0019]

[0020] Where k represents the sampling point of the RMS value in the previous resting stage, N represents the total sampling points of the RMS value in the previous resting stage, and PrRP i represents the pre-resting potential of channel i;

[0021] The expression of the former resting variability is as follows:

[0022]

[0023] Where, k represents the RMS sampling point, st represents the starting sampling point of the previous resting stage, ed represents the ending sampling point of the previous resting stage, mean represents the average myoelectric potential of the previous resting stage, and PrVRC i represents the pre-resting variability of channel i, PrVRC i The smaller it is, the more stable the neuromuscular control is;

[0024] Characteristics of the post-resting phase include post-resting potential and post-resting variability;

[0025] The calculation formula of the post-resting potential is as follows:

[0026]

[0027] Where k represents the sampling point of the RMS value in the post-resting phase, N represents the total sampling points of the RMS value in the post-resting phase, and PoRP i represents the post-resting potential of channel i;

[0028] The expression of the posterior resting variability is as follows:

[0029]

[0030] Where, k represents the RMS sampling point, st represents the starting sampling point of the post-resting phase, ed represents the ending sampling point of the post-resting phase, mean represents the average myoelectric potential of the post-resting phase, PoVRC i represents the post-resting variability of channel i, PoVRC i The smaller it is, the more stable the nerve's control over the muscle is;

[0031] The characteristics of the fast contraction and release phase include the fast contraction and release rise time, the maximum potential of the fast contraction, and the fast contraction and release recovery time;

[0032] The calculation formula of fast retraction and fast release rise time is as follows:

[0033]

[0034] Among them, t represents the actual time point of rapid contraction, t Max(RMSi) Indicates the time point of the maximum contraction value of fast contraction and fast release, t i Indicates the time point below the last resting threshold before fast playback, FRT i It represents the fast contraction and release rise time of channel i. The smaller the FRT, the stronger the innervation of fast-twitch type II fibers and the faster the nerve conduction velocity.

[0035] The formula for calculating the maximum potential of rapid contraction is as follows:

[0036]

[0037] Among them, n means that the pelvic floor muscles perform n times of rapid contraction and release, k means the kth time of rapid contraction and release, and max (RMSik) is the maximum contractile muscle strength in the kth fast contraction and release of channel i, RCP i is the n-time average maximum contraction muscle strength of the ith channel;

[0038] The formula for calculating the fast rewind and fast playback recovery time is as follows:

[0039]

[0040] Among them, t represents the actual time point of rapid contraction, t Max(RMSi) Indicates the time point of the maximum contraction value of fast contraction and fast release, t iIndicates the time point at which the first resting threshold is reached after fast playback and fast release. FFT i It represents the fast contraction and release recovery time of channel i. The smaller the FFT is, the easier it is for fast-twitch type II fibers to relax and less likely to hypertonic.

[0041] Characteristics of the contraction phase included sustained contraction potential, sustained contraction variability, sustained contraction rise time, and sustained contraction recovery time;

[0042] The sustained contraction potential is calculated as follows:

[0043]

[0044] Among them, n means that the pelvic floor muscles are contracted and maintained for n times, k means the kth contraction and maintenance, and RMS ik Maintain the RMS potential for the kth contraction of channel i, st k Indicates the starting sampling point of the k-th contraction hold, ed k Indicates the end sampling point of the kth contraction hold, TCP i represents the sustained contraction potential of channel i, TCP i The higher the value, the stronger the Class I muscle.

[0045] Sustained contraction variability was calculated as follows:

[0046]

[0047] Where n represents the pelvic floor muscle contraction for n times, k represents the RMS sampling point, j represents the jth sustained contraction, st represents the starting sampling point of the sustained contraction phase, ed represents the ending sampling point of the sustained contraction phase, mean represents the average electromyographic potential in the previous resting phase, and TVRC represents the average electromyographic potential in the previous resting phase. i represents the sustained contraction variability of channel i, TVRC i The smaller it is, the more stable the nerve's innervation of type I muscles is;

[0048] The calculation formula for the sustained contraction rise time is as follows:

[0049]

[0050] Among them, n means that the pelvic floor muscles are contracted and maintained for n times, t means the actual time point of continuous contraction, and t FThr(RMSij) represents the time point of the first plateau of the jth sustained contraction, t ij TRT represents the time point below the last resting threshold before the jth sustained contraction. i Indicates the sustained contraction rise time of channel i, TRT i The smaller the number, the stronger the innervation of fast-twitch type I fibers and the faster the nerve conduction velocity;

[0051] The calculation formula of sustained contraction recovery time is as follows:

[0052]

[0053] Among them, n represents the n-time contraction of the pelvic floor muscles, t represents the actual time point of continuous contraction, and t LThr(RMSij) represents the time point of the last plateau of the jth sustained contraction, t ij represents the time point below the first resting threshold after the jth sustained contraction, FRT i represents the sustained contraction recovery time of channel i, FRT i The smaller it is, the faster the type I muscle fibers are, the easier it is for the nerves to relax and not to become hypertonic.

[0054] The 1-minute holding phase characteristics included the 1-minute sustained contraction potential, the median frequency of the first 10 seconds, the median frequency of the last 10 seconds, the endurance contraction rate, and the variability of the 1-minute holding phase;

[0055] The formula for calculating the 1-minute sustained contraction potential is as follows:

[0056]

[0057] Among them, st represents the starting sampling point of 1-minute continuous contraction, ed represents the ending sampling point of 1-minute continuous contraction, and ECP i represents the 1-minute sustained contraction potential of the i channel, ECP i The stronger it is, the higher the Class I muscle;

[0058] The median frequency for the first 10 seconds is calculated as follows:

[0059]

[0060] Where k represents the frequency point in the electromyographic spectrum, P i represents the power spectrum of the myoelectricity of channel i, f1 represents the lower limit of the myoelectricity frequency bandwidth, f2 represents the upper limit of the myoelectricity frequency bandwidth, PoMF i Indicates the median frequency of channel i in the last 10 seconds, PrMF i represents the median frequency of the first 10 seconds of channel i;

[0061] The median frequency for the last 10 seconds is expressed as follows:

[0062]

[0063] Where k represents the frequency point in the EMG spectrum, Pi represents the EMG power spectrum of channel i, f1 represents the lower limit of the EMG frequency bandwidth, f2 represents the upper limit of the EMG frequency bandwidth, PoMFi represents the median frequency of channel i in the last 10 seconds, and PrMF irepresents the median frequency of the first 10 seconds of channel i;

[0064] The endurance contraction rate was calculated by calculating the ratio of the average EMG amplitude in the last 10 seconds of the 1-minute holding phase to the average EMG amplitude in the first 10 seconds. The calculation formula is as follows:

[0065]

[0066] Among them, ed represents the start time of the last 10 seconds, edd represents the end time of the last 10 seconds, st represents the start time of the first 10 seconds, std represents the end time of the first 10 seconds, REC i Represents the endurance shrinkage rate of channel i, REC i The closer it is to 1, the stronger the fatigue resistance is;

[0067] The expression of variability during the 1-minute hold phase is as follows:

[0068]

[0069] Where k represents the RMS sampling point, st represents the starting sampling point of the 1-minute holding phase, ed represents the ending sampling point of the 1-minute holding phase, mean represents the average electromyographic potential of the 1-minute holding phase, and EVRC i represents the 1-minute holding phase variability of channel i, EVRC i The smaller it is, the more stable the nerve's control over the muscle is;

[0070] Cough stage characteristics include pelvic floor muscle and abdominal muscle delay, cough stage delay asymmetry index, and cough stage muscle strength asymmetry index;

[0071] The delay formula for the pelvic floor muscles and abdominal muscles is as follows:

[0072] DT=Onset(RMSi)-Onset(Ab)

[0073] Among them, Onset represents the algorithm for finding the starting point of the signal generation, Ab represents the electromyographic signal of the abdominal muscles during the coughing stage, and DT i It indicates the time difference between the i-channel electromyography and the contraction of the abdominal muscles during coughing. Healthy people will contract the pelvic floor muscles in advance, and DT is negative or close to 0, while patients with stress urinary incontinence are generally positive.

[0074] The formula for the cough phase delay asymmetry index is as follows:

[0075]

[0076] Among them, DT iIt represents the time difference between the contraction of channel i electromyography and abdominal muscle during coughing. TAAI(x,y) represents the phase delay asymmetry index of coughing. The more x in TAAI(x,y) is distributed in the middle, the more coordinated the timing of neural reflex activity is. i represents the horizontal coordinate of the i-th channel on the electrode expansion surface, y i represents the ordinate of the i-th channel on the electrode expansion surface;

[0077] The formula for the muscle asymmetry index of cough stages is as follows:

[0078]

[0079] Among them, RMS i Indicates the myoelectric MRS of channel i during coughing, x i represents the horizontal coordinate of the i-th channel on the electrode expansion surface, y i It represents the ordinate of the i-th channel on the electrode expansion surface. PAAI(x,y) represents the muscle asymmetry index of the cough stage. The more x in PAAI(x,y) is distributed in the middle, the higher the coordination of the strength of the neural reflex activity.

[0080] Furthermore, the features include the ratio of resting potential before and after, the ratio of median frequency between the last 10 seconds and the first 10 seconds of the 1-minute holding phase;

[0081] The formula for calculating the ratio of the resting potential before and after is as follows:

[0082]

[0083] Among them, PRP i It represents the ratio of the resting potential of the i channel before and after, which indicates the change of muscle tension before and after contraction. The high tension of the pelvic floor muscles at rest may cause chronic pelvic pain and lead to fatigue of the pelvic floor muscles. The closer the PRP is to 1, the normal relaxation of the pelvic floor muscles after exercise. The PRP greater than 1 indicates that the muscles are more likely to be hypertonic after exercise.

[0084] Ratio of median frequency in the last 10 seconds of the 1-minute hold phase to the first 10 seconds:

[0085]

[0086] Among them, MFR i It represents the ratio of the last 10 seconds to the first 10 seconds of the 1-minute hold phase of channel i. The larger the MFR is greater than 1 or the closer it is to 1, the stronger the fatigue resistance of the pelvic floor muscles is.

[0087] Furthermore, step S3 reduces the dimension of the features of the pre-resting stage, the fast contraction and release stage, the contraction and hold stage, the 1-minute hold stage, the coughing stage, and the post-resting stage based on principal component analysis PCA and autoencoder technology, and then integrates the features through a deep learning model according to the formula:

[0088] ZX

[0089] Among them, Z is the eigenvector after dimensionality reduction, X is the original feature matrix of the six stages and between stages, and W is the eigenvector matrix;

[0090] h=σWx+be)

[0091]

[0092] Among them, h is the encoded feature vector, W e is the encoder weight matrix, W d is the weight matrix of the decoder, b e and b d is the bias term, σ is the activation function.

[0093] Furthermore, the existing pelvic floor feature database is used to match the reduced-dimensional features with the features in the database. The specific matching steps are as follows:

[0094] Step S4-1: The feature database uses a hash table to establish an index system;

[0095] h(k)=k mod m

[0096] Among them, h(k) is the hash value, k is the key value, m is the size of the hash table, and mod means remainder;

[0097] Step S4-2: using a machine learning model to classify the pelvic floor feature database in the database;

[0098]

[0099] Among them, f(x) is the classification result, α i is the Lagrange multiplier, y i is the category label, K is the kernel function, and b is the bias term;

[0100] Step S4-3: Match the reduced dimension features with the features in the pelvic floor feature database through an algorithm, and select the sample with the smallest distance d. The calculation formula of d is as follows:

[0101]

[0102] Where d(x,y) is the Euclidean distance between feature vectors x and y.

[0103] Furthermore, the pelvic floor electromyography diseases are diagnosed by calculating the similarity between the features after dimensionality reduction and the samples, using the phase relationship function for calculation:

[0104]

[0105] Among them, r is the correlation coefficient, xi and yi are the components of the two input feature vectors, and They are the means of the two input feature vectors respectively. The closer the r value is to 1, the more similar the two features are. The final diagnostic results include pelvic floor chronic pelvic pain positioning, pelvic floor hypertonia positioning, pelvic floor stress urinary incontinence, and reflex activity symmetry assessment, pelvic floor muscle strength assessment, weak muscle group assessment, pelvic floor fast and slow muscle innervation symmetry and other health assessment results.

[0106] Compared with the prior art, the present invention has the following beneficial effects:

[0107] 1. Use a high-density array vaginal pelvic floor probe, which has an additional channel for detecting abdominal muscle electromyography to ensure the completeness of the input information.

[0108] 2. Six pelvic floor function modes are adopted, with a total of 20 characteristics, including rest, fast and slow movements, and reflex activities.

[0109] 3. In order to solve the technical problem of dimensionality explosion, the deep learning method is used to reduce the dimension and extract the main features. At the same time, it also strengthens the correlation between the features, avoiding the existing method of only performing single-dimensional pelvic floor function on a single feature, and accurately analyzing more pelvic floor diseases.

[0110] 4. The high-density vaginal electrode with uniformly distributed electrodes is used, which facilitates practical operation and use, transfers the difficulty of analysis and positioning to the algorithm step, and improves the practicality and operability of the entire system. BRIEF DESCRIPTION OF THE DRAWINGS

[0111] Figure 1 This is a schematic diagram of a method for analyzing pelvic floor nerve and muscle group image data based on deep learning according to the present invention. DETAILED DESCRIPTION

[0112] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0113] Example: Figure 1As shown, the present invention provides a technical solution, a pelvic floor nerve muscle group image data analysis method based on deep learning, the analysis method comprising:

[0114] Step S1: using a vaginal pelvic floor probe to collect pelvic floor electromyography images;

[0115] The vaginal pelvic floor probe is attached with a channel for detecting abdominal muscle electromyography, and the electrode points are evenly distributed on the vaginal pelvic floor probe;

[0116] In an embodiment of the present invention, a high-density array vaginal pelvic floor probe is used, preferably 32+1 channels or 64+1 channels, wherein an additional channel is used to detect abdominal muscle electromyography, and the electrode points are evenly distributed on the probe.

[0117] Step S2: collecting sample points from the pelvic floor electromyography image, and performing feature extraction on the pelvic floor electromyography image;

[0118] Among them, step S2 is to collect sample points from the pelvic floor electromyography image with a window of 50ms and a step size of 10ms, and calculate the root mean square of the electromyography signal in each channel of the original pelvic floor electromyography and the abdominal muscle channel according to the sample points. The calculation formula is as follows:

[0119]

[0120] Among them, i represents the number of channels, k represents the sampling point, and x i (k) represents the kth sampling point of the original electromyographic signal of the i-th channel, N represents the total sampling points in the window, RMS i represents the root mean square of the electromyographic signal in channel i;

[0121] The pelvic floor electromyography image includes six stages, which are: pre-resting stage, fast contraction and release stage, contraction stage, 1-minute holding stage, coughing stage and post-resting stage. The six stages are feature extracted based on the root mean square.

[0122] In an embodiment of the present invention, the pelvic floor electromyography image is collected in the following manner:

[0123] In the pre-resting phase, the pelvic floor muscles remain at rest for 1 minute;

[0124] In the fast contraction and release phase, the pelvic floor muscles are contracted and released 5 times with a 10-second interval between each movement.

[0125] During the contraction and holding phase, the pelvic floor muscles are contracted and held 5 times, each time for 10 seconds and with a 10-second interval between each time.

[0126] During the 1-minute hold phase, the pelvic floor muscles are held for 1 minute.

[0127] During the coughing phase, the patient coughed five times, with 10 seconds between each cough;

[0128] In the post-resting phase, the pelvic floor muscles remain at rest for 1 minute;

[0129] Among them, the characteristics of the pre-resting stage include pre-resting potential and pre-resting variability;

[0130] The formula for calculating the front resting potential is as follows:

[0131]

[0132] Where k represents the sampling point of the RMS value in the previous resting stage, N represents the total sampling points of the RMS value in the previous resting stage, and PrRP i represents the pre-resting potential of channel i;

[0133] The expression of the former resting variability is as follows:

[0134]

[0135] Where, k represents the RMS sampling point, st represents the starting sampling point of the previous resting stage, ed represents the ending sampling point of the previous resting stage, mean represents the average myoelectric potential of the previous resting stage, and PrVRC i represents the pre-resting variability of channel i;

[0136] Among them, the characteristics of the post-resting phase include post-resting potential and post-resting variability;

[0137] The calculation formula of the post-resting potential is as follows:

[0138]

[0139] Where k represents the sampling point of the RMS value in the post-resting phase, N represents the total sampling points of the RMS value in the post-resting phase, and PoRP i represents the post-resting potential of channel i;

[0140] The expression of the posterior resting variability is as follows:

[0141]

[0142]

[0143] Where, k represents the RMS sampling point, st represents the starting sampling point of the post-resting phase, ed represents the ending sampling point of the post-resting phase, mean represents the average myoelectric potential of the post-resting phase, PoVRC i represents the post-resting variability of channel i;

[0144] Among them, the characteristics of the fast-retraction and fast-release stage include the fast-retraction and fast-release rise time, the maximum potential of the fast contraction, and the fast-retraction and fast-release recovery time;

[0145] The calculation formula of fast retraction and fast release rise time is as follows:

[0146]

[0147] Among them, t represents the actual time point of rapid contraction, t Max(RMSi) Indicates the time point of the maximum contraction value of fast contraction and fast release, t i Indicates the time point below the last resting threshold before fast playback, FRT i It represents the fast-release rise time of channel i;

[0148] The formula for calculating the maximum potential of rapid contraction is as follows:

[0149]

[0150] Among them, n means that the pelvic floor muscles perform n times of rapid contraction and release, k means the kth time of rapid contraction and release, and max (RMSik) is the maximum contractile muscle strength in the kth fast contraction and release of channel i, RCP i is the n-time average maximum contraction muscle strength of channel i;

[0151] The formula for calculating the fast rewind and fast playback recovery time is as follows:

[0152]

[0153] Among them, t represents the actual time point of rapid contraction, t Max(RMSi) Indicates the time point of the maximum contraction value of fast contraction and fast release, t i Indicates the time point at which the first resting threshold is reached after fast playback and fast release. FFT i Indicates the fast-retraction and fast-release recovery time of channel i;

[0154] Among them, the characteristics of the contraction phase include sustained contraction potential, sustained contraction variability, sustained contraction rise time, and sustained contraction recovery time;

[0155] The sustained contraction potential is calculated as follows:

[0156]

[0157] Among them, n means that the pelvic floor muscles are contracted and maintained for n times, k means the kth contraction and maintenance, and RMS ik Maintain the RMS potential for the kth contraction of channel i, st k Indicates the starting sampling point of the k-th contraction hold, ed k Indicates the end sampling point of the kth contraction hold, TCP irepresents the sustained contraction potential of channel i;

[0158] Sustained contraction variability was calculated as follows:

[0159]

[0160] Where n represents the pelvic floor muscle contraction for n times, k represents the RMS sampling point, j represents the jth sustained contraction, st represents the starting sampling point of the sustained contraction phase, ed represents the ending sampling point of the sustained contraction phase, mean represents the average electromyographic potential in the previous resting phase, and TVRC represents the average electromyographic potential in the previous resting phase. i represents the sustained contraction variability of channel i;

[0161] The calculation formula for the sustained contraction rise time is as follows:

[0162]

[0163] Among them, n means that the pelvic floor muscles are contracted and maintained for n times, t means the actual time point of continuous contraction, and t FThr(RMSij) represents the time point of the first plateau of the jth sustained contraction, t ij TRT represents the time point below the last resting threshold before the jth sustained contraction. i It represents the sustained contraction rise time of channel i;

[0164] The calculation formula of sustained contraction recovery time is as follows:

[0165]

[0166] Among them, n represents the n-time contraction of the pelvic floor muscles, t represents the actual time point of continuous contraction, and t LThr(RMSij) represents the time point of the last plateau of the jth sustained contraction, t ij represents the time point below the first resting threshold after the jth sustained contraction, FRT i represents the sustained contraction recovery time of channel i;

[0167] Among them, the characteristics of the 1-minute holding phase include the 1-minute sustained contraction potential, the median frequency of the first 10 seconds, the median frequency of the last 10 seconds, the endurance contraction rate, and the variability of the 1-minute holding phase;

[0168] The formula for calculating the 1-minute sustained contraction potential is as follows:

[0169]

[0170] Among them, st represents the starting sampling point of 1-minute continuous contraction, ed represents the ending sampling point of 1-minute continuous contraction, and ECP i represents the 1-minute sustained contraction potential of channel i;

[0171] The median frequency for the first 10 seconds is calculated as follows:

[0172]

[0173] Where k represents the frequency point in the electromyographic spectrum, P i represents the power spectrum of the myoelectricity of channel i, f1 represents the lower limit of the myoelectricity frequency bandwidth, f2 represents the upper limit of the myoelectricity frequency bandwidth, PoMF i Indicates the median frequency of channel i in the last 10 seconds, PrMF i represents the median frequency of the first 10 seconds of channel i;

[0174] The median frequency for the last 10 seconds is expressed as follows:

[0175]

[0176] Where k represents the frequency point in the electromyographic spectrum, P i represents the power spectrum of the myoelectricity of channel i, f1 represents the lower limit of the myoelectricity frequency bandwidth, f2 represents the upper limit of the myoelectricity frequency bandwidth, PoMFi represents the median frequency of channel i in the last 10 seconds, PrMF i represents the median frequency of the first 10 seconds of channel i;

[0177] The endurance contraction rate is calculated by calculating the ratio of the average EMG amplitude in the last 10 seconds of the 1-minute holding phase to the average EMG amplitude in the first 10 seconds. The calculation formula is as follows:

[0178]

[0179] Among them, ed represents the start time of the last 10 seconds, edd represents the end time of the last 10 seconds, st represents the start time of the first 10 seconds, std represents the end time of the first 10 seconds, REC i It represents the endurance contraction rate of channel i;

[0180] The expression of variability during the 1-minute hold phase is as follows:

[0181]

[0182] Where k represents the RMS sampling point, st represents the starting sampling point of the 1-minute holding phase, ed represents the ending sampling point of the 1-minute holding phase, mean represents the average electromyographic potential of the 1-minute holding phase, and EVRC i represents the 1-min holding phase variability of channel i;

[0183] Among them, the characteristics of cough phase include pelvic floor muscle and abdominal muscle delay, cough phase delay asymmetry index, and cough phase muscle strength asymmetry index;

[0184] The delay formula for the pelvic floor muscles and abdominal muscles is as follows:

[0185] DT=Onset(RMSi)-Onset(Ab)

[0186] Among them, Onset represents the algorithm for finding the starting point of the signal generation, Ab represents the electromyographic signal of the abdominal muscles during the coughing stage, and DT i It indicates the time difference between the electromyography of channel i and the contraction of the abdominal muscles during coughing;

[0187] The formula for the cough phase delay asymmetry index is as follows:

[0188]

[0189] Among them, DT i represents the time difference between the contraction of channel i electromyography and abdominal muscle during coughing, TAAI(x,y) represents the phase delay asymmetry index of coughing, x i represents the horizontal coordinate of the i-th channel on the electrode expansion surface, y i represents the ordinate of the i-th channel on the electrode expansion surface;

[0190] The formula for the muscle asymmetry index of the cough stage is as follows:

[0191]

[0192] Among them, RMS i Indicates the myoelectric MRS of channel i during coughing, x i represents the horizontal coordinate of the i-th channel on the electrode expansion surface, y i represents the ordinate of the ith channel on the electrode expansion surface, and PAAI(x,y) represents the muscle asymmetry index of the cough stage.

[0193] Among them, the features also included the ratio of resting potential before and after, the ratio of median frequency in the last 10 seconds to the first 10 seconds of the 1-minute holding phase;

[0194] The formula for calculating the ratio of the resting potential before and after is as follows:

[0195]

[0196] Among them, PRP i represents the ratio of the before and after resting potentials of channel i;

[0197] The median frequency ratio of the last 10 seconds to the first 10 seconds of the 1-minute hold phase was calculated as follows:

[0198]

[0199] Among them, MFR iIt represents the ratio of the last 10 seconds to the first 10 seconds of the 1-minute holding phase of channel i.

[0200] Step S3: reduce the dimension of features and integrate the features;

[0201] In an embodiment of the present invention, step S3 reduces the dimension of the features of the pre-resting stage, the fast contraction and release stage, the contraction and hold stage, the 1-minute hold stage, the coughing stage, and the post-resting stage based on principal component analysis PCA and autoencoder technology, and then integrates the features through a recursive neural network RNN, according to the formula:

[0202] ZX

[0203] Among them, Z is the eigenvector after dimensionality reduction, X is the original feature matrix of the six stages and between stages, and W is the eigenvector matrix;

[0204] h=σ(Wx+be

[0205]

[0206] Among them, h is the encoded feature vector, W e is the encoder weight matrix, W d is the weight matrix of the decoder, b e and b d is the bias term, σ is the activation function.

[0207] Step S4: using the existing pelvic floor feature database, matching the reduced dimension features with the features in the database;

[0208] In an embodiment of the present invention, the existing pelvic floor feature database is used to match the features after dimensionality reduction with the features in the database. The specific matching steps are as follows:

[0209] Step S4-1: The feature database uses a hash table to establish an index system;

[0210] h(k)=k mod m

[0211] Among them, h(k) is the hash value, k is the key value, and m is the size of the hash table;

[0212] Step S4-2: using a support vector machine (SVM) to classify the pelvic floor feature database in the database;

[0213]

[0214] Among them, f(x) is the classification result, αi is the Lagrange multiplier, yi is the category label, K is the kernel function, and b is the bias term;

[0215] Step S4-3: Use the KNN nearest neighbor algorithm to match the high-density pelvic floor muscle electrical energy features after dimensionality reduction with the data in the pelvic floor feature database, and select the sample with the smallest distance d. The calculation formula of d is as follows:

[0216]

[0217] Where d(x,y) is the Euclidean distance between feature vectors x and y.

[0218] Step S5: diagnose pelvic floor electromyography diseases by calculating the similarity between the features after dimension reduction and the samples;

[0219] Among them, the diagnosis of pelvic floor electromyography diseases is carried out by calculating the similarity between the features after dimensionality reduction and the samples, and using the phase relationship function for calculation:

[0220]

[0221] Among them, r is the correlation coefficient, xi and yi are the components of the two input feature vectors, and are the means of the two input eigenvectors respectively.

[0222] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. 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. A pelvic floor nerve muscle group image data analysis method based on deep learning, characterized in that: The analysis method comprises: Step S1: using a vaginal pelvic floor probe to collect pelvic floor electromyography images; Step S2: collecting sample points from the pelvic floor electromyography image, and performing feature extraction on the pelvic floor electromyography image; Step S3: reduce the dimension of features and integrate the features; Step S4: using the existing pelvic floor feature database, matching the reduced dimension features with the features in the database; Step S5: Diagnose pelvic floor electromyography diseases by calculating the similarity between the features after dimensionality reduction and the samples.

2. The method for analyzing pelvic floor nerve muscle group image data based on deep learning according to claim 1, characterized in that: The vaginal pelvic floor probe is attached with a channel for detecting abdominal muscle myoelectricity, and the electrode points are evenly distributed on the vaginal pelvic floor probe.

3. The method for analyzing pelvic floor nerve muscle group image data based on deep learning according to claim 1, characterized in that: The step S2 is to collect sample points from the pelvic floor electromyography image with a window of 50 ms and a step length of 10 ms, and calculate the root mean square of the electromyography signal in each channel of the original pelvic floor electromyography and the abdominal muscle channel according to the sample points. The calculation formula is as follows: Among them, i represents the number of channels, k represents the sampling point, and x i (k) represents the kth sampling point of the original electromyographic signal of the i-th channel, N represents the total sampling points in the window, RMS i Represents the RMS value of the electromyographic signal in channel i.

4. The method for analyzing pelvic floor nerve muscle group image data based on deep learning according to claim 1, characterized in that: The pelvic floor electromyography image includes six stages, which are: a pre-resting stage, a fast contraction and release stage, a contraction stage, a 1-minute holding stage, a coughing stage, and a post-resting stage, and features of the six stages are extracted based on the root mean square. The characteristics of the pre-resting phase include pre-resting potential and pre-resting variability; The front resting potential calculation formula is as follows: Where k represents the sampling point of the RMS value in the previous resting stage, N represents the total sampling points of the RMS value in the previous resting stage, and PrRP i represents the pre-resting potential of channel i; The former resting variability expression is as follows: Where, k represents the RMS sampling point, st represents the starting sampling point of the previous resting stage, ed represents the ending sampling point of the previous resting stage, mean represents the average myoelectric potential of the previous resting stage, and PrVRC i represents the pre-resting variability of channel i; The characteristics of the post-resting phase include post-resting potential and post-resting variability; The calculation formula of the post-resting potential is as follows: Where k represents the sampling point of the RMS value in the post-resting phase, N represents the total sampling points of the RMS value in the post-resting phase, and PoRP i represents the post-resting potential of channel i; The expression of the posterior resting variability is as follows: Where, k represents the RMS sampling point, st represents the starting sampling point of the post-resting phase, ed represents the ending sampling point of the post-resting phase, mean represents the average myoelectric potential of the post-resting phase, PoVRC i represents the post-resting variability of channel i; The characteristics of the fast-retraction and fast-release stage include the fast-retraction and fast-release rise time, the fast-contraction maximum potential, and the fast-retraction and fast-release recovery time; The calculation formula for the fast-release and fast-retraction rise time is as follows: Among them, t represents the actual time point of rapid contraction, t Max(RMSi) Indicates the time point of the maximum contraction value of fast contraction and fast release, t i Indicates the time point below the last resting threshold before fast playback, FRT i It represents the fast-release rise time of channel i; The formula for calculating the maximum potential of rapid contraction is as follows: Among them, n means that the pelvic floor muscles perform n times of rapid contraction and release, k means the kth time of rapid contraction and release, and max (RMSik) is the maximum contractile muscle strength in the kth fast contraction and release of channel i, RCP i is the n-time average maximum contraction muscle strength of channel i; The fast rewind and fast play recovery time calculation formula is as follows: Among them, t represents the actual time point of rapid contraction, t Max(RMSi) Indicates the time point of the maximum contraction value of fast contraction and fast release, t i Indicates the time point at which the resting threshold is first reached after fast playback and fast release. FFT i Indicates the fast-retraction and fast-release recovery time of channel i; The characteristics of the contraction phase include sustained contraction potential, sustained contraction variability, sustained contraction rise time, and sustained contraction recovery time; The sustained contraction potential calculation formula is as follows: Among them, n means that the pelvic floor muscles are contracted and maintained for n times, k means the kth contraction and maintenance, and RMS ik Maintain the RMS potential for the kth contraction of channel i, st k Indicates the starting sampling point of the k-th contraction hold, ed k Indicates the end sampling point of the kth contraction hold, TCP i represents the sustained contraction potential of channel i; The sustained contraction variability was calculated as follows: Where n represents the pelvic floor muscle contraction for n times, k represents the RMS sampling point, j represents the jth sustained contraction, st represents the starting sampling point of the sustained contraction phase, ed represents the ending sampling point of the sustained contraction phase, mean represents the average electromyographic potential in the previous resting phase, and TVRC represents the average electromyographic potential in the previous resting phase. i represents the sustained contraction variability of channel i; The calculation formula for the sustained contraction rise time is as follows: Among them, n means that the pelvic floor muscles are contracted and maintained for n times, t means the actual time point of continuous contraction, and t FThr(RMSij) represents the time point of the first plateau of the jth sustained contraction, t ij TRT represents the time point below the last resting threshold before the jth sustained contraction. i It represents the sustained contraction rise time of channel i; The calculation formula for the sustained contraction recovery time is as follows: Among them, n represents the n-time contraction of the pelvic floor muscles, t represents the actual time point of continuous contraction, and t LThr(RMSij) represents the time point of the last plateau of the jth sustained contraction, t ij represents the time point below the first resting threshold after the jth sustained contraction, FRT i represents the sustained contraction recovery time of channel i; The characteristics of the 1-minute holding phase include 1-minute sustained contraction potential, median frequency of the first 10 seconds, median frequency of the last 10 seconds, endurance contraction rate, and 1-minute holding phase variability; The 1-minute sustained contraction potential calculation formula is as follows: Among them, st represents the starting sampling point of 1-minute continuous contraction, ed represents the ending sampling point of 1-minute continuous contraction, and ECP i represents the 1-minute sustained contraction potential of channel i; The median frequency calculation formula for the first 10 seconds is as follows: Where k represents the frequency point in the electromyographic spectrum, P i represents the power spectrum of the myoelectricity of channel i, f1 represents the lower limit of the myoelectricity frequency bandwidth, f2 represents the upper limit of the myoelectricity frequency bandwidth, PoMF i Indicates the median frequency of channel i in the last 10 seconds, PrMF i represents the median frequency of the first 10 seconds of channel i; The median frequency of the last 10 seconds is expressed as follows: Where k represents the frequency point in the electromyographic spectrum, P i represents the power spectrum of the myoelectricity of channel i, f1 represents the lower limit of the myoelectricity frequency bandwidth, f2 represents the upper limit of the myoelectricity frequency bandwidth, PoMF i Indicates the median frequency of channel i in the last 10 seconds, PrMF i represents the median frequency of the first 10 seconds of channel i; The endurance contraction rate is calculated by calculating the ratio of the average electromyographic amplitude in the last 10 seconds of the 1-minute holding phase to the average electromyographic amplitude in the first 10 seconds. The calculation formula is as follows: Among them, ed represents the start time of the last 10 seconds, edd represents the end time of the last 10 seconds, st represents the start time of the first 10 seconds, std represents the end time of the first 10 seconds, REC i It represents the endurance contraction rate of channel i; The 1-minute hold phase variability expression is as follows: Where, k represents the RMS sampling point, st represents the starting sampling point of the 1-minute holding stage, ed represents the ending sampling point of the 1-minute holding stage, mean represents the average electromyographic potential of the 1-minute holding stage, and EVRC i represents the 1-min holding phase variability of channel i; The characteristics of the coughing stage include pelvic floor muscle and abdominal muscle delay, coughing stage delay asymmetry index, and coughing stage muscle strength asymmetry index; The pelvic floor muscle and abdominal muscle delay formula is as follows: DT i =Onset(RMS i )-Onset(Ab); Among them, Onset represents the algorithm for finding the starting point of the signal generation, Ab represents the electromyographic signal of the abdominal muscles during the coughing stage, and DT i It indicates the time difference between the electromyography of channel i and the contraction of the abdominal muscles during coughing; The cough phase delay asymmetry index formula is as follows: Among them, DT i represents the time difference between the contraction of channel i electromyography and abdominal muscle during coughing, TAAI(x,y) represents the phase delay asymmetry index of coughing, x i represents the horizontal coordinate of the i-th channel on the electrode expansion surface, y i represents the ordinate of the i-th channel on the electrode expansion surface; The formula of the muscle asymmetry index of the cough stage is as follows: Among them, RMS i represents the myoelectric MRS of channel i during coughing, and PAAI(x,y) represents the muscle asymmetry index of the coughing stage.

5. The method for analyzing pelvic floor nerve muscle group image data based on deep learning according to claim 4, characterized in that: The characteristics also include the ratio of the front and back resting potential PRP i , MFR of the last 10 seconds of the 1-minute hold phase to the first 10 seconds i .

6. The method for analyzing pelvic floor nerve muscle group image data based on deep learning according to claim 1, characterized in that: The step S3 reduces the dimension of the features of the pre-resting stage, the fast contraction and release stage, the contraction and hold stage, the 1-minute hold stage, the coughing stage, and the post-resting stage based on principal component analysis PCA and autoencoder technology, and then integrates the features through a deep learning model according to the formula: Z = XW; Among them, Z is the eigenvector after dimensionality reduction, X is the original feature matrix of the six stages and between stages, and W is the eigenvector matrix; h=σ(W e x+b e ) Among them, h is the encoded feature vector, W e is the encoder weight matrix, W d is the weight matrix of the decoder, b e and b d is the bias term and σ is the activation function.

7. The method for analyzing pelvic floor nerve muscle group image data based on deep learning according to claim 1, characterized in that: Using the existing pelvic floor feature database, the reduced-dimensional features are matched with the features in the database. The specific matching steps are as follows: Step S4-1: The feature database uses a hash table to establish an index system; h(k)=k mod m Among them, h(k) is the hash value, k is the key value, m is the size of the hash table, and mod means remainder; Step S4-2: using a machine learning model to classify the pelvic floor feature database in the database; Among them, f(x) is the classification result, αi is the Lagrange multiplier, yi is the category label, K is the kernel function, and b is the bias term; Step S4-3: Use an algorithm to match the features after dimension reduction with the features in the pelvic floor feature database, and select the sample with the smallest distance d. The calculation formula of d is as follows: Where d(x,y) is the Euclidean distance between feature vectors x and y.

8. The method for analyzing pelvic floor nerve muscle group image data based on deep learning according to claim 1, characterized in that: Diagnose pelvic floor electromyography diseases by calculating the similarity between the reduced-dimensional features and the samples using a phase relationship function: Among them, r is the correlation coefficient, xi and yi are the components of the two input feature vectors, and are the means of the two input eigenvectors respectively.

Images