CN114611563B - Method for identifying neurogenic damage of pin-pole electromyogram across parts - Google Patents
Method for identifying neurogenic damage of pin-pole electromyogram across parts Download PDFInfo
- Publication number
- CN114611563B CN114611563B CN202210525623.5A CN202210525623A CN114611563B CN 114611563 B CN114611563 B CN 114611563B CN 202210525623 A CN202210525623 A CN 202210525623A CN 114611563 B CN114611563 B CN 114611563B
- Authority
- CN
- China
- Prior art keywords
- data
- electromyogram
- pin
- pole
- classifier
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/24—Detecting, measuring or recording bioelectric or biomagnetic signals of the body or parts thereof
- A61B5/316—Modalities, i.e. specific diagnostic methods
- A61B5/389—Electromyography [EMG]
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7225—Details of analog processing, e.g. isolation amplifier, gain or sensitivity adjustment, filtering, baseline or drift compensation
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Abstract
The invention relates to a method for identifying neurogenic damage of a needle pole electromyogram across a part, which comprises the following steps: firstly, carrying out certain pretreatment on an original electromyographic signal input into a system; secondly, extracting the features of the preprocessed pin-pole electromyogram data, and cascading the extracted features to form a feature vector(ii) a Thirdly, carrying out electromyogram characteristic data of the needle pole of a certain partInputting the data into an integrated classifier for integrated learning, and implementing integrated learning on output results of a plurality of classifiers by adopting a weighted integration strategy to implement prediction output of all sample data of each individual; the fourth step, in the classification stage, the prediction probabilities of a plurality of classifiers are determinedWeighted summation is carried out to obtain the final prediction probabilityJudging whether the patient is ill or not by taking 0.5 as a threshold value, thereby realizing classification of the electromyographic signals; wherein, w i Is the probability weight of each classifier.
Description
Technical Field
The invention relates to the technical field of auxiliary medical treatment, in particular to a method for identifying needle pole electromyographic neurogenic damage in a cross-site mode.
Background
The pin-pole electromyogram is used as a common detection tool for clinical diagnosis of nervous system diseases and orthopedic diseases, and the functional states of peripheral nerves, neurons, neuromuscular junctions and muscles are further determined by acquiring intramuscular data of a subject under different muscle contraction conditions, checking nerve and muscle excitation and conduction functions and the like, so that the analysis on specific parts and damage degree of neuromuscular diseases is realized, and a decisive clinical reference is provided for the determination of disease diagnosis and treatment schemes. Years of clinical practice shows that the pin-pole electromyography examination has important clinical significance for the diagnosis, prognosis evaluation and detection of neuromuscular diseases. However, clinical pin electromyogram data analysis can only rely on experienced neurology needle electromyogram specialist physicians. The manual quantitative analysis-based mode has a large amount of deviation and instability, which causes obstacles to the early diagnosis of nervous system diseases and surgery-related diseases, and the related technology needs to be solved urgently.
During muscle contraction, the motor unit generates excitatory impulses, i.e. motor unit action potentials, under the control of the central nervous system. The electromyographic signals can be decomposed into a plurality of action electric potential sequences of the movement units, and the time and shape information of the movement units is extracted through quantitative analysis of the electric potentials of the movement units, so that a clinician is helped to diagnose various neuromuscular diseases, and better understanding of the neuromuscular system under different disease states is facilitated. At present, the clinical needle electromyogram decomposition is mainly based on a manual quantitative analysis method, and the needle electromyogram signals are quantitatively analyzed by observing with naked eyes and manually measuring the duration and the amplitude of the action potential of movement. The electromyographic signals of each muscle are influenced by traditional analysis, a good standard value needs to be determined, but the electromyographic signals are influenced by the race, the specific part of the muscle, the nutrition degree and even the measured environmental temperature, and the estimated and calibrated standard value cannot accurately reflect the specific condition to a certain individual, so that errors cannot be avoided during manual alignment.
In addition, since only 2-6 points of motion unit can be captured by the pin-polar electromyogram each time, the muscles are required to be repeatedly detected before the diagnosis conclusion is obtained, and the average material for each muscle is required to be obtained 3-6 times, for example, in the examination method of the pin-polar electromyogram to the right brachial triceps, an examination physician needs to perform multiple needle insertions (3-4 times) in different directions, generally, data needs to be collected for multiple times in each direction of the muscle, and the total number of the insertions is 5-30 steps (times), which brings pain to the patient. The electromyographic signals obtained by the pin-pole electromyography are influenced by measuring techniques, measuring parts, skin thickness, matching of patients and the like, and certain fluctuation can be generated, which also influences manual interpretation.
In current clinical work, the needle pole electromyogram decomposition is mainly based on an artificial quantitative analysis method, the duration and the amplitude of a movement action potential are measured by visual observation and computer assistance workers to carry out quantitative analysis on needle pole electromyogram signals, and the qualitative of the neurogenic damage is carried out by comparing measured values with normal values and calculating differences with specific values.
In summary, the following main problems exist in the manual quantitative analysis method in the prior art:
1. the traditional manual quantitative analysis method of the needle electrode electromyogram needs to be calibrated by depending on normal values, the normal values of muscles at different parts are different, but the normal value of the measured muscle is influenced by the environment, the state of a tested person and the skill of an operator and has volatility, so the calibrated normal value has certain deviation, and the diagnosis accuracy is influenced to a certain degree.
2. Because the normal values of the muscles at different parts are different, the operator needs to compare the normal values one by one according to clinical experience, compare the measured values with the normal values, and calculate the difference through the sum ratio to carry out the characterization of the neurogenic damage. The multiple data of each muscle need comparison calculation, and a certain time is needed for calculation. The clinical analysis of one patient test (the upper and lower limbs, bilateral symmetry muscle sample 4 common affected part muscles) is completed on average, which usually needs 30 minutes to 1 hour. This greatly limits the number of cases that a clinician can perform daily, limiting the mass development of clinical procedures.
3. Because of the limitation of the traditional acupuncture electromyography method, a clinician needs to sample a muscle part for multiple times in order to avoid the influence of manipulations, environment and patient coordination, and the maximum amplitude and the maximum time limit obtained during sampling are taken as operation standards, so that multiple comparisons are needed, and the patient often suffers from repeated needle insertion and multiple puncture on the sampling part of the patient.
Disclosure of Invention
The invention aims to provide a method for identifying neurogenic damage of a needle pole electromyogram across parts, which at least solves the technical problem of how to realize the learning of neurogenic damage across individuals and across parts by comprehensively analyzing all needle pole electromyogram sample data of the same part or across parts of each individual, so that the experimental design and the analysis process are closer to the analysis process of a clinician.
The invention aims to construct an integrated learning framework for cross-individual data learning, and realize multi-transform domain and multi-scale representation of pin-pole electromyogram signals by extracting linear and nonlinear omics characteristics; through a weighted integration classification scheme constructed by multiple classifiers, the identification deviation of a single classifier is reduced, and the classification performance of cross-individual and cross-part feature data is improved; through automatic feature extraction and a clustering learning framework, workload and misjudgment rate of manual interpretation are reduced, accuracy and working efficiency of pin-polar electromyogram data analysis are improved, and computer-aided diagnosis performance of the pin-polar electromyogram data on neurogenic damage is further improved.
In order to achieve the purpose, the invention provides a method for identifying the neurogenic damage of the needle pole electromyogram in a cross-site manner, which comprises the following steps:
firstly, carrying out certain pretreatment on an original electromyographic signal input into a system;
secondly, extracting the features of the preprocessed pin-pole electromyogram data, and cascading the extracted features to form a feature vector;
Thirdly, carrying out electromyogram characteristic data of the needle pole of a certain partInputting the data into an integrated classifier for integrated learning, and implementing integrated learning on output results of a plurality of classifiers by adopting a weighted integration strategy to implement prediction output of all sample data of each individual;
the fourth step, in the classification stage, the prediction probabilities of a plurality of classifiers are determinedWeighted summation is carried out to obtain the final prediction probabilityJudging whether the patient is ill or not by taking 0.5 as a threshold value, thereby realizing the classification of the electromyographic signals;
wherein, w i Is the probability weight of each classifier.
Preferably, the pretreatment in the first step specifically comprises:
1) for each individual raw electromyographic signalPerforming a window division process in whichi=1,2,……,N,When dividing the window, the length is selected asThen, 50% of windows are overlapped to prevent the data extraction problem caused by data deviation; forming preliminary sample data after window division;
2) performing baseline drift removal operation on the preliminary sample data through curve fitting;
3) the data from baseline wander was normalized.
Preferably, the second step of extracting the features of the preprocessed pin-pole electromyogram data specifically includes:
extracting various statistical characteristics of the time domain waveform of the pin-pole electromyogram data in the time domain;
in a frequency domain, extracting the frequency spectrum characteristic and the power spectrum characteristic of the pin-pole electromyogram frequency spectrum after Fourier transform;
in the wavelet domain, performing wavelet decomposition and wavelet packet decomposition on the pin-pole electromyogram data by adopting a dB4 wavelet, and extracting corresponding statistical characteristics and energy characteristics of the decomposition coefficients;
in the aspect of nonlinear characteristics, a multi-fractal detrending fluctuation analysis method is adopted to extract the nonlinear characteristics of the pin-electrode electromyogram data.
Further preferably, the various statistical characteristics of the time domain waveform of the extracted pin-pole electromyogram data include first-order, second-order and third-order statistical characteristics.
Preferably, the integrated classifier in the third step is formed by a plurality of single classifiers.
Further preferably, the plurality of single classifiers include LDA, Adaboost, DecisionTree, RandomForest, Catboost and XGboost.
Preferably, the implementation of the ensemble learning of the output results of the plurality of classifiers in the third step specifically includes:
1) 80% of the selected characteristic dataAs input data, training as training data of the selected plurality of classifiers respectively;
2) after each classifier independently trains the training data, the remaining 20 percent of feature data is used for carrying out single-part classification effect test, the feature data of the other part is used for carrying out cross-part classification effect test, and the prediction probability of each classifier is storedAnd sample accuracy;
3) Sample accuracy for each classifierThe following processing is carried out to obtain the related accuracy rate a of each classifier i ,
Wherein, A min Represents the minimum of the sample accuracy of all classifiers;
A max represents the maximum value of sample accuracy for all classifiers;
meanwhile, normalization processing is carried out on the related accuracy to obtain the probability weight w of each classifier i The formula is as follows:
advantageous effects
Compared with the prior art, the invention has the beneficial effects that:
according to the method for identifying the neurogenic damage of the pin-pole electromyogram at the cross-site, the similarity of a training set and a test set is reduced in a mode of dividing the pin-pole electromyogram data according to individuals, and the method is different from a learning method of dividing the pin-pole electromyogram data according to samples in the conventional pin-pole electromyogram data analysis, so that the cross-individual analysis of the pin-pole electromyogram data is realized; and through cross-site analysis, the extraction of the common characteristic of the neurogenic injury is realized.
The method for identifying the neurogenic damage of the pin-pole electromyogram across the part adopts the combination of the advanced computer processing technology and the medical diagnosis, can greatly improve the efficiency and the accuracy of quantitative analysis of the pin-pole electromyogram data, enables the diagnosis result to be more accurate, and achieves the purposes of early discovery, early diagnosis and early treatment of neuromuscular diseases.
Drawings
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention.
FIG. 1 is a flow chart of the algorithm of the present invention.
Detailed Description
The present invention is described in more detail below to facilitate an understanding of the present invention.
In order to reduce the workload and the misjudgment rate of manual interpretation, improve the accuracy and the working efficiency of analyzing the pin-pole electromyogram data and further improve the computer-aided diagnosis performance of the pin-pole electromyogram data on the neurogenic injury, the inventor of the invention designs the following experiments.
First, experimental data
The data used in the experiment originated from the third hospital of Beijing university. The experiment totally comprises a data set of 3 muscle parts, a first interosseous muscle data set from third hospital of Beijing university, the age of the study subject is distributed from 24 years to 80 years, and the study subject comprises 81 neurogenic injury individuals and the pin-pole electromyogram data of 52 normal control individuals; the age of the study subjects ranged from 23 to 80 years from the tibialis anterior data set from the third hospital of Beijing university, including the pin-polar electromyogram data of 56 neurogenic injury individuals and 54 normal control individuals. In this experiment, five-fold cross-over random experiments were performed on all training data and test data. When random experiments are carried out on the pin-pole electromyogram data of a single muscle part, 80% of pin-pole electromyogram individual characteristic data are selected as training data, and the rest data are used as tests.
Second, experiment software environment
In this embodiment, the system used is an Ubuntu 18.04 system, the GPU is configured as a NVIDIA GeForce 1080Ti 11 × 8G, and the development software environment is a server of python 3.7.
Third, experimental method and flow
The invention aims to construct an integrated learning framework for cross-individual data learning, and realize multi-transform domain and multi-scale representation of pin-pole electromyogram signals by extracting linear and nonlinear omics characteristics; through a weighted integration classification scheme constructed by multiple classifiers, the identification deviation of a single classifier is reduced, and the classification performance of cross-individual and cross-part feature data is improved; through automatic feature extraction and a clustering learning framework, workload and misjudgment rate of manual interpretation are reduced, accuracy and working efficiency of pin-polar electromyogram data analysis are improved, and computer-aided diagnosis performance of the pin-polar electromyogram data on neurogenic damage is further improved.
Based on the above experiments, the invention provides a method for identifying the neurogenic damage of the needle pole electromyogram in a cross-site manner, which comprises the following steps:
firstly, certain preprocessing is carried out on an original electromyographic signal input into a system. The method comprises the following specific steps:
1) for each individual raw electromyographic signal(i=1,2,……,N) Performing window division processing, wherein the length is selected asThen there is 50% overlap between the windows to prevent situations where data skew causes data extraction problems. Forming preliminary sample data after window division;
2) performing baseline drift removal operation on the preliminary sample data through curve fitting;
3) carrying out standardization processing on the data of baseline drift removal;
secondly, extracting the features of the preprocessed pin-pole electromyogram data, and cascading the extracted features to form a feature vector. The method comprises the following specific steps:
extracting various statistical characteristics of the time domain waveform of the pin-pole electromyogram data in a time domain, wherein the statistical characteristics comprise first-order, second-order and third-order statistical characteristics;
in a frequency domain, extracting the frequency spectrum characteristic and the power spectrum characteristic of the pin pole electromyogram frequency spectrum after Fourier transform;
in the wavelet domain, for the pin-pole electromyogram data, a dB4 wavelet is adopted to carry out wavelet decomposition and wavelet packet decomposition, and corresponding statistical characteristics and energy characteristics of decomposition coefficients are extracted.
In the aspect of nonlinear characteristics, a multi-fractal detrending fluctuation analysis method is adopted to extract the nonlinear characteristics of the pin-electrode electromyogram data.
Thirdly, carrying out electromyogram characteristic data of the needle pole of a certain partAnd inputting the data into an ensemble classifier for ensemble learning. The ensemble classifier is formed by collecting a plurality of single classifiers (the classifiers adopted in the embodiment are LDA, Adaboost, DecisionTree, RandomForest, castboost and XGboost), and the ensemble learning of the output results of the plurality of classifiers is realized by adopting a weighted ensemble strategy, so that the predicted output of all sample data of each individual is realized. The method comprises the following steps:
1) 80% of the selected characteristic dataAs input data, training as training data of the selected plurality of classifiers respectively;
2) after each classifier independently trains the training data, the remaining 20 percent of feature data is used for carrying out single-part classification effect test, the other part of feature data is used for carrying out cross-part classification effect test, and the prediction probability of each classifier is storedAnd sample accuracy;
3) Sample accuracy for each classifierThe following processing is carried out to obtain the related accuracy rate a of each classifier i ,
Wherein A is min Represents the minimum of the sample accuracy of all classifiers;
A max represents the maximum value of sample accuracy for all classifiers;
meanwhile, normalization processing is carried out on the related accuracy to obtain the probability weight w of each classifier i The formula is as follows:
the fourth step, in the classification stage, the prediction probabilities of a plurality of classifiers are determinedWeighted summation is carried out to obtain the final prediction probabilityAnd judging whether the patient is ill or not by taking 0.5 as a threshold value, thereby realizing the classification of the electromyographic signals.
The algorithm flow chart of the present invention is shown in fig. 1.
Fourth, experimental results
The first interosseous muscle and the tibialis anterior muscle are respectively used as training sets one by one, and the test effect of the method on the trans-individual and trans-site pin-pole electromyography data is analyzed.
As shown in table 1, which is a comparison of the accuracy of the classification samples of the average integration and the weighted integration, it can be seen that the weighted integration shows better classification effect than the average integration.
TABLE 1 comparison of sample accuracy for cross-individual, cross-site classification of weighted integration and average integration
TABLE 2 comparison of the first interosseous muscle training, tibialis anterior muscle testing single classifier classification effect
As shown in table 2, which shows the comparison of the classification effect of the single classifier during the first intercondylar training and the tibialis anterior testing, it can be seen that the best prediction effect is obtained by the castboost compared to the other single classifiers. Based on this result, the present application performed comparative analysis of the method of the present invention with Catboost as follows.
As shown in tables 3-5, compared with a single classifier, the method of the present invention has the advantages that the classification accuracy, sensitivity and specificity are improved by 1% to 7% in the cross-individual and single-site test; in the test of crossing individuals and crossing parts, the classification accuracy, the sensitivity and the specificity are improved by 2 to 5 percent; compared with the test of a single part, the cross-part test also achieves better experimental effect, and the omics characteristics can effectively reflect the description of the common characteristics of the neurogenic injury.
TABLE 3 Cross-Individual, Cross-site Classification Individual accuracy comparison
TABLE 4 Cross-Individual, Cross-site Classification sensitivity comparison
TABLE 5 Cross-Individual, Cross-site Classification specificity comparison
In a comprehensive way, in the cross-individual and cross-site pin polar electromyogram data classification experiment, compared with a single classifier in the traditional machine learning, the method disclosed by the invention classifies the electromyogram signals through an integrated learning framework on the basis of pre-extracted pin polar electromyogram feature data, so that the accuracy of pin polar electromyogram neuron damage detection is improved; the addition of the weight coefficient enables the classification effect to be more stable compared with the traditional machine learning, and the stability is improved while the efficiency is ensured.
The method is used for effectively extracting the characteristics based on the needle electrode electromyogram data of the early-stage neuromuscular disease with damage, so that the extraction and analysis of the abnormal characteristics of the early-stage damage of the neurogenic damage of the muscle are realized. The method is based on the research work of a machine learning identification classification algorithm, simultaneously considers the linear and nonlinear omics characteristics of the extracted pin-pole electromyogram data, constructs a weighted integration learning framework, realizes the early identification and classification of damaged pin-pole electromyogram data, realizes the cross-site method for identifying neurogenic damage by pin-pole electromyogram, improves the accuracy of early diagnosis of damaged neuromuscular diseases, and provides auxiliary reference for clinical diagnosis.
The invention discloses a method for identifying neurogenic damage of a needle pole electromyogram in a cross-site manner, which is characterized in that the method comprises the following two aspects:
firstly, algorithm aspect:
(1) the similarity of a training set and a test set is reduced by dividing the pin-polar electromyogram data according to individuals, and the cross-individual analysis of the pin-polar electromyogram data is realized by being different from a learning method in the past in which the pin-polar electromyogram data is divided according to samples;
(2) extracting the common characteristic of the neurogenic injury through cross-site analysis;
(3) by extracting linear and nonlinear omics characteristics, multi-transform domain and multi-scale representation of the pin-pole electromyogram signals is realized, and detection and extraction of the neurogenic injury common characteristics of different parts are effectively realized;
(4) through a weighted integration classification scheme constructed by multiple classifiers, the identification deviation of a single classifier is reduced, and the classification performance of cross-individual and cross-part feature data is improved;
(5) through automatic feature acquisition and classification learning frames, workload and misjudgment rate of manual interpretation are reduced, accuracy and working efficiency of analysis of the pin polar electromyogram data are improved, and computer-aided diagnosis performance of the pin polar electromyogram data on neurogenic injury is further improved.
Secondly, in the functional aspect:
(1) the judgment standard is extracted according to the electromyogram characteristics of the neurogenic damage without depending on a standard value, the electrophysiological substance is better met, and a constant error is avoided.
(2) When the analysis mode of the invention does not require clinical examination any more, the doctor pursues the maximum detection value during operation, and only needs to standardize and finish the examination process, thereby greatly reducing the examination time and the pain of the patient, ensuring the detection effect, simultaneously making the electromyogram examination more easily accepted, and obtaining the examination result more quickly.
The preferred embodiments of the present invention have been described in detail with reference to the accompanying drawings, however, the present invention is not limited to the specific details of the above embodiments, and various simple modifications can be made to the technical solution of the present invention within the technical idea of the present invention, and these simple modifications are within the protective scope of the present invention.
It should be noted that the various technical features described in the above embodiments can be combined in any suitable manner without contradiction, and the invention is not described in any way for the possible combinations in order to avoid unnecessary repetition.
In addition, any combination of the various embodiments of the present invention is also possible, and the same should be considered as the disclosure of the present invention as long as it does not depart from the spirit of the present invention.
The foregoing describes preferred embodiments of the present invention, but is not intended to limit the invention thereto. Modifications and variations of the embodiments disclosed herein may be made by those skilled in the art without departing from the scope and spirit of the invention.
Claims (5)
1. A method for identifying a needle pole electromyographic neurogenic damage in a cross-region manner is characterized by comprising the following steps of:
firstly, carrying out certain pretreatment on an original electromyographic signal input into a system;
secondly, extracting the features of the preprocessed pin-pole electromyogram data, and cascading the extracted features to form a feature vector;
Thirdly, carrying out electromyogram characteristic data of the needle pole of a certain partInputting the data into an integrated classifier for integrated learning, and implementing integrated learning on output results of a plurality of classifiers by adopting a weighted integration strategy to implement prediction output of all sample data of each individual;
the fourth step, in the classification stage, the prediction probabilities of a plurality of classifiers are determinedWeighted summation is carried out to obtain the final prediction probabilityJudging whether the patient is ill or not by taking 0.5 as a threshold value, thereby realizing classification of the electromyographic signals;
wherein, w i A probability weight for each classifier;
the pretreatment in the first step specifically comprises:
1) for each individual raw electromyographic signalPerforming a window division process in whichi=1,2,……,N,When dividing the window, the length is selected asThen, 50% of windows are overlapped to form initial sample data;
2) performing baseline drift removal operation on the preliminary sample data through curve fitting;
3) carrying out standardization processing on the data of baseline drift removal;
the implementation of ensemble learning of the output results of the plurality of classifiers in the third step specifically includes:
1) 80% of the selected characteristic dataAs input data, training as training data of the selected plurality of classifiers respectively;
2) after each classifier independently trains the training data, the remaining 20 percent of feature data is used for carrying out single-part classification effect test, the feature data of the other part is used for carrying out cross-part classification effect test, and the prediction probability of each classifier is storedAnd sample accuracy;
3) Sample accuracy for each classifierThe following processing is carried out to obtain the related accuracy rate a of each classifier i ,
Wherein A is min Represents the minimum of the sample accuracy of all classifiers;
A max represents the maximum value of sample accuracy for all classifiers;
meanwhile, normalization processing is carried out on the related accuracy to obtain the probability weight w of each classifier i The formula is as follows:
2. the method for identifying the neurogenic impairment of the pin-pole electromyogram across the site according to claim 1, wherein the second step of performing feature extraction on the preprocessed pin-pole electromyogram data specifically comprises:
extracting various statistical characteristics of the time domain waveform of the pin-pole electromyogram data in the time domain;
in a frequency domain, extracting the frequency spectrum characteristic and the power spectrum characteristic of the pin-pole electromyogram frequency spectrum after Fourier transform;
in the wavelet domain, performing wavelet decomposition and wavelet packet decomposition on the pin-pole electromyogram data by adopting a dB4 wavelet, and extracting corresponding statistical characteristics and energy characteristics of the decomposition coefficients;
in the aspect of nonlinear characteristics, a multi-fractal detrending fluctuation analysis method is adopted to extract the nonlinear characteristics of the pin-electrode electromyogram data.
3. The method for identifying the neurogenic impairment of the pin-polar electromyogram across the site according to claim 2, wherein the various statistical features of the time-domain waveform of the extracted pin-polar electromyogram data include first, second and third order statistical features.
4. The method for identifying neurogenic impairment of pin-polar electromyogram according to claim 1, wherein the integrated classifier in the third step is assembled from a plurality of single classifiers.
5. The method of identifying neurogenic impairment of a pin-polar electromyogram across a site of claim 4, wherein the plurality of single classifiers comprises LDA, Adaboost, decisionTree, RandomForest, Catboost, and XGBoost.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210525623.5A CN114611563B (en) | 2022-05-16 | 2022-05-16 | Method for identifying neurogenic damage of pin-pole electromyogram across parts |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210525623.5A CN114611563B (en) | 2022-05-16 | 2022-05-16 | Method for identifying neurogenic damage of pin-pole electromyogram across parts |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114611563A CN114611563A (en) | 2022-06-10 |
CN114611563B true CN114611563B (en) | 2022-08-23 |
Family
ID=81870489
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210525623.5A Active CN114611563B (en) | 2022-05-16 | 2022-05-16 | Method for identifying neurogenic damage of pin-pole electromyogram across parts |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114611563B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117198463A (en) * | 2023-11-06 | 2023-12-08 | 中国科学院大学 | Method for identifying needle electrode electromyography motor neuron damage by crossing data fields |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130253375A1 (en) * | 2012-03-21 | 2013-09-26 | Henry Nardus Dreifus | Automated Method Of Detecting Neuromuscular Performance And Comparative Measurement Of Health Factors |
CN110598628B (en) * | 2019-09-11 | 2022-08-02 | 南京邮电大学 | Electromyographic signal hand motion recognition method based on integrated deep learning |
CN110598676B (en) * | 2019-09-25 | 2022-08-02 | 南京邮电大学 | Deep learning gesture electromyographic signal identification method based on confidence score model |
CN112861604B (en) * | 2020-12-25 | 2022-09-06 | 中国科学技术大学 | Myoelectric action recognition and control method irrelevant to user |
CN114266270A (en) * | 2021-11-22 | 2022-04-01 | 南京航空航天大学 | Electromyographic signal decoding method based on recurrent neural network and ensemble learning |
-
2022
- 2022-05-16 CN CN202210525623.5A patent/CN114611563B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN114611563A (en) | 2022-06-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Bairy et al. | Automated diagnosis of depression electroencephalograph signals using linear prediction coding and higher order spectra features | |
US20110087125A1 (en) | System and method for pain monitoring at the point-of-care | |
US20150245800A1 (en) | Method for Detection Of An Abnormal Sleep Pattern In A Person | |
Benazzouz et al. | EMG Feature selection for diagnosis of neuromuscular disorders | |
CN114611563B (en) | Method for identifying neurogenic damage of pin-pole electromyogram across parts | |
Bozkurt et al. | Comparison of AR parametric methods with subspace-based methods for EMG signal classification using stand-alone and merged neural network models | |
Ganguly et al. | A non-invasive approach for fetal arrhythmia detection and classification from ecg signals | |
KR20190126604A (en) | A device for diagnosis of peripheral neuropathy using wavelet transform of needle electromyography signal | |
Kempfner et al. | Automatic REM sleep detection associated with idiopathic rem sleep Behavior Disorder | |
Chen et al. | Automatic multichannel intramuscular electromyogram decomposition: progressive fastICA peel-off and performance validation | |
Antunes et al. | A morphology-based feature set for automated Amyotrophic Lateral Sclerosis diagnosis on surface electromyography | |
CN113425298A (en) | Method for analyzing depression degree by collecting data through wearable equipment | |
Lu et al. | Pulse waveform analysis for pregnancy diagnosis based on machine learning | |
Helal et al. | A hybrid approach for artifacts removal from EEG recordings | |
Istenic et al. | Classification of neuromuscular disorders using wavelets and entropy metrics on surface electromyograms | |
Chandel et al. | Patient specific seizure onset-offset latency detection using long-term EEG signals | |
Fadav et al. | A machine learning approach for addiction detection using phase amplitude coupling of EEG signals | |
Athavale et al. | Classification of periodic leg movements through actigraphy signal analysis | |
Achmamad et al. | Fast automatic detection of amyotrophic lateral sclerosis disease based on Euclidean distance metric | |
Sarossy et al. | Relationship between glaucoma and complexity measures of the electroretinogram | |
Williams et al. | Quantitative assessment of electroencephalogram reactivity in comatose patients on extracorporeal membrane oxygenation | |
Biswas et al. | Automated System Design for the Identification of Sleep Disorder: Cross-correlation and SVM Based Approach | |
CN117257303B (en) | Anxiety detection method and device, electronic equipment and storage medium | |
Chowdhury et al. | A Machine Learning Approach for Person Authentication from EEG Signals | |
Pattichis et al. | Introduction to this special issue: intelligent data analysis in electromyography and electroneurography |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |