Electromyogram F wave classification analysis system and method based on support vector machine
Through the F-wave classification analysis system of electromyography based on support vector machine, the problems of large labor expenditure and classification difficulties in F-wave data analysis are solved, and the automatic classification and efficient analysis of F-waves are realized, which significantly improves accuracy and efficiency.
Patent Information
- Application Number
- CN202510327284.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-19
- Publication Date
- 2025-06-13
AI Technical Summary
In the prior art, data analysis after F wave output requires a lot of manpower, and it is difficult to classify the F wave curve, and there are problems of inaccurate judgment and mixed bias.
The electromyography F-wave classification analysis system based on support vector machine is adopted, including training modules and classification modules. The training sample data is obtained through the data bus, the Lagrange multiplier, classification threshold and support vector are calculated, and the fitting regression of nonlinear data is completed using the kernel method to realize the automatic classification of F waves.
The F-wave extraction accuracy is significantly improved, the noise resistance is enhanced, the method adaptability is improved, the multi-dimensional characteristics of the F-wave are comprehensively extracted, the classification accuracy and stability are significantly improved, and the F-wave analysis is fully automated, with efficiency improvements of more than 95%.
Smart Images
Figure CN120131041A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and particularly to an electromyogram F-wave classification and analysis system and method based on a support vector machine. Background Art
[0002] The F-wave often appears after the compound muscle action potential and is a delayed motor potential, which is one of the late motor responses obtained by conventional electromyogram research. The generation mechanism of the F-wave is that after peripheral nerve stimulation, some motor neurons are reversely activated, causing the muscle fibers to contract again to generate a potential. It usually refers to the action potential generated when the muscle nerve fibers are re-excited after a rest period after the first depolarization. This potential is usually slower than the initial M-wave (the action potential generated by a single muscle stimulation), so it is called the F-wave. The F-wave has important clinical value in evaluating motor nerve conduction, diagnosing peripheral nerve lesions, and monitoring the progression of nervous system diseases. The recording of the F-wave usually requires first obtaining the maximum amplitude value of the M-response, and then increasing the stimulation intensity to the supra-maximal stimulation level. The F-wave is often recorded using surface electrodes with at least 20 supra-maximal stimulations. The F-wave is often obtained by stimulating nerves such as the median nerve, ulnar nerve, and tibial nerve.
[0003] The F-wave can be used to evaluate nerve conduction velocity and muscle excitability. By analyzing the characteristic curve and latency of the F-wave, axonal degeneration, demyelinating degeneration, and demyelination combined with axonal degeneration can be further refined. Through statistical analysis of the software, the results show a consistent trend of change to evaluate the degree of nerve root injury and further for evaluating the therapeutic effect.
[0004] However, for the analysis of the data after the F-wave output, it takes a lot of manpower to statistically analyze and data-label the F-wave curve. It is difficult to classify different types of curves among the 20 F-wave curves output each time. At the same time, it is difficult to obtain the original data file of the curve, and only the output data result pictures can be used for manual visual observation and classification, resulting in problems of inaccurate judgment and confounding bias. Summary of the Invention
[0005] The purpose of the present invention is to solve the problems existing in the prior art, and propose an electromyogram F-wave classification and analysis system and method based on a support vector machine.
[0006] To achieve the above purpose, the present invention adopts the following technical solutions: An electromyogram F-wave classification and analysis system based on a support vector machine, including a training module and a classification module; The training module is used for: Obtaining training sample data from the outside through a data bus; And splitting the training sample data into sample data and sample labels; Calculate and output the Lagrange multipliers, classification thresholds, and support vectors; The classification module is used for: Obtain the Lagrange multipliers, classification thresholds, and support vectors required for classification, obtain the sample data to be classified from the outside through the data bus, and classify the samples according to the classification function.
[0007] Further, the training module includes a sample splitting unit, a sample label memory, a sample data memory, an inner product calculation unit, an inner product memory, and a Lagrange multiplier and classifier threshold calculation unit; The two output terminals of the sample splitting unit are respectively connected to the sample label memory and the sample data memory. The first output terminals of the sample label memory and the sample data memory are connected to the two input terminals of the inner product calculation unit. The output terminal of the inner product calculation unit is connected to the inner product memory. The second output terminals of the sample label memory and the sample data memory and the output terminal of the inner product memory are respectively connected to the H input terminals of the Lagrange multiplier and classifier threshold calculation unit. The output terminals of the Lagrange multiplier and classifier threshold calculation unit output H parameters including the Lagrange multipliers, classification thresholds, and support vectors.
[0008] Further, the sample splitting unit is composed of a first data distributor. The input terminal of the first data distributor receives the sample data, and the output terminals are connected to the sample data memory and the sample label memory.
[0009] Further, the inner product calculation unit includes a second data distributor, a third data distributor, a first multiplier, a second multiplier, and an adder. The sample data memory is connected to the input terminal of the second data distributor. The output terminal of the second data distributor is connected to the first multiplier. The first multiplier is connected to the first adder. The first adder is connected to the third data distributor. The output terminal of the third data distributor is connected to the second multiplier. The output of the second multiplier is connected to the inner product memory.
[0010] Further, the Lagrange multiplier and classifier threshold calculation unit includes a third multiplier, a fourth data distributor, a data selector, a subtractor, an AND gate, a first comparator, two second comparators, a first adder / subtractor, a second adder / subtractor, a second adder, and a third adder.
[0011] Further, the inner product memory is connected to the input end of the Lagrange multiplier and classification threshold calculation unit. The result output by the Lagrange multiplier and classification threshold calculation unit sequentially passes through a third multiplier, a second adder, and a fourth data distributor. One output end of the fourth data distributor is returned to the previous second adder, and the other output end is output to the H-th adder. The output end of the H-th adder is connected to a subtractor. The other input end of the subtractor is connected to the sample label memory. The output end of the subtractor is connected to a first comparator and a second comparator. The output ends of the first comparator and the second comparator are connected to an AND gate. The output end of the AND gate is connected to a first adder-subtractor, a second adder-subtractor, and a data selector to control the adjustment of the classifier parameters.
[0012] Further, the sample label memory, the sample data memory, and the inner product memory are respectively composed of registers for storing sample labels, sample data, and inner product calculation results.
[0013] According to a second aspect of the present disclosure, there is provided an electromyogram F-wave classification and analysis method based on a support vector machine, which uses an electromyogram F-wave classification and analysis system based on a support vector machine, including the following steps: Obtain training F-wave sample data from the outside; Detect, identify, and analyze the F-wave. If the waveform and text content are correct, record the waveform data; if the waveform is invalid and / or the text content is incorrect, use OCR to identify the text content or manually mark the data, and record the waveform data; After preprocessing the F-wave data, the training sample preparation function calculates the Lagrange multiplier and the classification threshold and outputs the results; Use the SVM to classify the F-wave according to the classification threshold and output the required waveform.
[0014] According to a third aspect of the present disclosure, there is provided an electronic device, including: At least one processor; A memory communicatively connected to the at least one processor; Wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to claim 8.
[0015] According to a fourth aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the method according to claim 8.
[0016] Compared with the prior art, the beneficial effects of the present invention are: The present invention performs normalization processing on measured data and divides it into K groups for cross-training, then imports the training data into the model, introduces genetic algorithm and particle swarm algorithm to adaptively search for the initial parameters of the SVR model, finally uses the Lagrange multiplier dual method to solve the "support vectors", uses the kernel method to complete the fitting regression of non-linear data, and obtains the regression result through cross-validation; 1. Significantly improve the accuracy of F-wave extraction: Through adaptive signal preprocessing and dynamic F-wave recognition algorithm, the recognition accuracy of F-wave is increased by 15 - 20%, and the misrecognition rate is reduced by more than 40%.
[0017] 2. Enhance the anti-noise ability: The multi-resolution wavelet analysis combined with the adaptive threshold function enables the method to maintain stable performance in a complex noise environment, and the average signal-to-noise ratio is increased by 13 - 16 dB.
[0018] 3. Improve the adaptability of the method: The dynamic parameter adjustment mechanism enables the method to adapt to different nerves and shows excellent generalization ability.
[0019] 4. Comprehensive feature extraction: The comprehensive feature extraction system captures the multi-dimensional features of F-wave, provides rich information for subsequent analysis, and reduces the average error of feature extraction to less than 3%.
[0020] 5. Optimize the classification performance: The multi-model integrated classification system significantly improves the classification accuracy and stability, and the average classification accuracy on multiple data sets is increased by 5 - 8%.
[0021] 6. Improve the analysis efficiency: Achieve the full automation of F-wave analysis, shorten the traditional manual analysis time from an average of 30 minutes per case to less than 1 minute, and the efficiency is increased by more than 95%. Description of the Drawings
[0022] Figure 1 It is a schematic diagram of the principle of the electromyogram F-wave classification and analysis system based on support vector machine proposed by the present invention; Figure 2 It is an example diagram of the waveform image of the electromyogram F-wave classification and analysis method based on support vector machine proposed by the present invention; Figure 3 It is an example diagram of data extraction and output of the electromyogram F-wave classification and analysis method based on support vector machine proposed by the present invention. Detailed Embodiments
[0023] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments.
[0024] Refer to Figures 1-3, An electromyogram F-wave classification and analysis system based on a support vector machine, including a training module and a classification module; The training module is used for: Obtain training sample data from the outside through a data bus; And split the training sample data into sample data and sample labels; Calculate and output Lagrange multipliers, classification thresholds, and support vectors; The classification module is used for: Obtain the Lagrange multipliers, classification thresholds, and support vectors required for classification, obtain the sample data to be classified from the outside through a data bus, and classify the samples according to the classification function.
[0025] In this embodiment, the training module includes a sample splitting unit, a sample label memory, a sample data memory, an inner product calculation unit, an inner product memory, and a Lagrange multiplier and classifier threshold calculation unit; The two output terminals of the sample splitting unit are respectively connected to the sample label storage and the sample data memory. The first output terminals of the sample label memory and the sample data memory are connected to the two input terminals of the inner product calculation unit. The output terminal of the inner product calculation unit is connected to the inner product memory. The second output terminals of the sample label memory and the sample data memory and the output terminal of the inner product memory are respectively connected to the H input terminals of the Lagrange multiplier and classifier threshold calculation unit. The output terminals of this Lagrange multiplier and classifier threshold calculation unit respectively output three parameters: Lagrange multipliers, classification thresholds, and support vectors.
[0026] In this embodiment, the sample splitting unit is composed of a first data distributor. The input terminal of the first data distributor receives sample data, and the output terminals are connected to the sample data memory and the sample label memory.
[0027] In this embodiment, the inner product calculation unit includes a second data distributor, a third data distributor, a first multiplier, a second multiplier, and an adder. The sample data memory is connected to the input terminal of the second data distributor. The output terminal of the second data distributor is connected to the first multiplier. The first multiplier is connected to the first adder. The first adder is connected to the third data distributor. The output terminal of the third data distributor is connected to the second multiplier. The output of the second multiplier is connected to the inner product memory.
[0028] In this embodiment, the Lagrange multiplier and classifier threshold calculation unit includes a third multiplier, a fourth data distributor, a data selector, a subtractor, an AND gate, a first comparator, two second comparators, a first addition and subtraction unit, a second addition and subtraction unit, a second adder, and a third adder.
[0029] In this embodiment, the inner product memory is connected to the input end of the Lagrange multiplier and classification threshold calculation unit. The results output by the Lagrange multiplier and classification threshold calculation unit sequentially pass through a third multiplier, a second adder, and a fourth data distributor. One output end of the fourth data distributor is returned to the previous second adder, and the other output end is output to the H-th adder. The output end of the H-th adder is connected to a subtractor. The other input end of the subtractor is connected to the sample label memory. The output end of the subtractor is connected to a first comparator and a second comparator. The output ends of the first comparator and the second comparator are connected to an AND gate. The output end of the AND gate is connected to a first adder / subtractor, a second adder / subtractor, and a data selector to control the adjustment of the classifier parameters.
[0030] In this embodiment, the sample label memory, the sample data memory, and the inner product memory are respectively composed of registers and are used to store the sample labels, sample data, and inner product calculation results.
[0031] According to the second aspect of the present disclosure, there is provided an electromyogram F-wave classification and analysis method based on a support vector machine, which adopts an electromyogram F-wave classification and analysis system based on a support vector machine, including the following steps: Obtain training sample data, that is, F-wave sample data, from the outside through a data bus; And split the training sample data into sample data and sample labels, and store them in the designated sample data memory and sample label memory respectively; Read the corresponding data of the designated sample from the sample data memory and the sample label memory, and store the obtained calculation results in the designated inner product memory; That is, detect, identify, and analyze the F-wave. If the waveform and text content are correct, record the waveform data; if the waveform is invalid and / or the text content is incorrect, use OCR to identify the text content or manually perform data marking and record the waveform data; The classification analysis system and method use the OCR recognition method to obtain the text content, and then output the corresponding recognition results to the training module for data processing according to the waveform and text content, and then output to the classification module to classify the waveforms with different clinical significances and output, so as to realize the rapid recognition and data processing of the waveforms under the condition that the original waveform data cannot be directly obtained or the waveform is defective, effectively improving the working efficiency of users, and providing a new solution for realizing rapid evaluation of motor nerve conduction, diagnosing peripheral nerve diseases, and monitoring nervous system diseases through electromyogram F-waves; Read the required relevant data from the sample data memory, the sample label memory, and the inner product memory, and calculate the Lagrange multiplier, the classification threshold, and the support vector; In the analysis system, that is, after preprocessing the F-wave data, the training sample preparation function calculates the Lagrange multipliers and the classification threshold, and outputs the results; According to the Lagrange multipliers, the classification threshold, and the support vectors, the sample data to be classified is obtained from the outside through the data bus, and the samples are classified according to the classification function.
[0032] In the classification module, that is, the F-wave is classified using SVM according to the classification threshold, and the required waveform is output.
[0033] According to a third aspect of the present disclosure, there is provided an electronic device, including: At least one processor; A memory communicatively connected to the at least one processor; Wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to claim 8.
[0034] According to a fourth aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the method according to claim 8.
[0035] Here, it should be noted that in this embodiment: 1. Data preprocessing function: The normalization process uses the following formula: Where x is the original data point, and xmin and xmax are the minimum and maximum values in the dataset respectively.
[0036] 2. Training sample preparation function: The formula for calculating the inner product is: Where xi and xj are the sample data, and n is the dimension of the features.
[0037] 3. Lagrange multiplier and classification threshold calculation function: The Lagrange multiplier αi is obtained by solving the following optimization problem: Where yi is the sample label, K(xi,xj) is the inner product, and m is the number of samples.
[0038] The classification threshold b is calculated by the following formula: Where xi is the support vector, and yi is the label of the corresponding support vector.
[0039] 4. Support vector calculation function: The support vectors are those samples xi that satisfy αi>0.
[0040] 5. Classification function: The classification function uses the following formula: where x is the sample to be classified, K(xi,x) is the inner product, and αi and b are parameters obtained through training.
[0041] 6. Cross-validation function: Cross-validation usually uses the following steps: Divide the dataset into k subsets.
[0042] For each subset, use it as the test set and the remaining subsets as the training set.
[0043] Calculate the accuracy rate on the test set.
[0044] Take the average accuracy rate of all subsets as the final evaluation result.
[0045] The sample splitting unit splits the training sample data into sample data and sample labels, and stores them in the specified sample data memory and sample label memory respectively. The inner product calculation unit reads the corresponding data of the specified sample from the sample data memory and the sample label memory respectively, and stores the obtained calculation result in the specified inner product memory. The lagrange multiplier and classification threshold calculation unit reads the required relevant data from the sample data memory, the sample label memory, and the inner product memory, and calculates the lagrange multiplier and the classification threshold. The classification module is connected to the training module to obtain the lagrange multiplier, classification threshold, and support vectors required for classification. The classification unit obtains the sample data to be classified from the outside through the data bus, reads the required relevant parameters from the memory, and finally classifies the sample according to the classification function.
[0046] As mentioned above, it is only the preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.
Claims
1. The electromyography F wave classification and analysis system based on support vector machine is characterized by: An electromyography F wave classification and analysis system based on support vector machine, including a training module and a classification module; The training module is used to: Obtain training sample data from the outside through the data bus; And split the training sample data into sample data and sample labels; Calculate the output lagrange multiplier, classification threshold and support vector; The classification module is used to: Obtain the lagrange multiplier, classification threshold and support vector required for classification, obtain the sample data to be classified from the outside through the data bus, and classify the samples according to the classification function.
2. The electromyography F wave classification and analysis system based on support vector machine according to claim 1, characterized in that: The training module includes a sample splitting unit, a sample label memory, a sample data memory, an inner product calculation unit, an inner product memory and a lagrange multiplier and a classifier width calculation unit; The two output ends of the sample splitting unit are respectively connected to the sample label storage and the sample data storage, the first output ends of the sample label storage and the sample data storage are connected to the two input ends of the inner product calculation unit, the output end of the inner product calculation unit is connected to the inner product storage, the second output ends of the sample label storage and the sample data storage and the output end of the inner product storage are respectively connected to the H input ends of the lagrange multiplier and classification width value calculation unit, and the output ends of the lagrange multiplier and classification width value calculation unit respectively output the lagrange multiplier, the classification width, and the H parameters of the support vector.
3. The electromyography F wave classification and analysis system based on support vector machine according to claim 2, characterized in that: The sample splitting unit is composed of a first data distributor, an input end of which receives sample data, and an output end of which is connected to a sample data storage device and a sample label storage device.
4. The electromyography F wave classification and analysis system based on support vector machine according to claim 3, characterized in that: The inner product calculation unit includes a second data distributor and a third data distributor, a first multiplier, a second multiplier and an adder, the sample data memory is connected to the input end of the second data distributor, the output end of the second data distributor is connected to the first multiplier, the first multiplier is connected to the first adder, the first adder is connected to the third data distributor, the output end of the third data distributor is connected to the second multiplier, and the output of the second multiplier is connected to the inner product memory.
5. The electromyography F wave classification and analysis system based on support vector machine according to claim 4, characterized in that: The lagrange multiplier and classification width calculation unit includes a Hth multiplier, a fourth data distributor, a data selector, a subtractor, an AND port, a first comparator, two second comparators, a first adder-subtractor, a second adder-subtractor, a second adder and a Hth adder.
6. The electromyography F wave classification and analysis system based on support vector machine according to claim 5, characterized in that: The inner product memory is connected to the input end of the lagrange multiplier and the classification width calculation unit, and the results output by the lagrange multiplier and the classification width calculation unit are sequentially passed through the third multiplier, the second adder and the fourth data distributor, wherein one output end of the fourth data distributor is returned to the previous second adder, and the other output end is output to the Hth adder, and the output end of the Hth adder is connected to the subtractor, and the other input end of the subtractor is connected to the sample label memory, and the output end of the subtractor is connected to the first comparator and the second comparator, and the output ends of the first comparator and the second comparator are connected to the AND port, and the output end of the AND port is connected to the first adder-subtractor, the second adder-subtractor and the data selector to control the adjustment of the classifier parameters.
7. The electromyography F wave classification and analysis system based on support vector machine according to claim 6, characterized in that: The sample label memory, sample data memory and inner product memory are respectively composed of registers, and are used to store data of sample labels, sample data and inner product calculation results.
8. An electromyogram F wave classification and analysis method based on a support vector machine, using an electromyogram F wave classification and analysis system based on a support vector machine according to any one of claims 1 to 7, characterized in that: The following steps are involved: Obtain training F wave sample data from outside; Detect, identify and analyze the F wave. If the waveform and text content are correct, record the waveform data. If the waveform is invalid and / or the text content is wrong, use OCR to identify the text content or manually mark the data and record the waveform data. After preprocessing the F wave data, the sample preparation function is trained, the Lagrange multiplier and classification threshold are calculated, and the results are output; Use SVM to classify F waves according to the classification threshold and output the required waveform.
9. An electronic device, characterized in that: include: at least one processor; a memory communicatively coupled to the at least one processor; The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method of claim 8.
10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to cause the computer to execute the method according to claim 8.