Neural signal waveform classification method, computer device and storage medium

By using multiple preset waveform statistical indicators and unsupervised learning models in neural signal waveform classification combined with template library verification, the problem of inconsistency and accuracy of neural signal waveform classification in the existing technology is solved, and more efficient and Accurate classification of action potentials and noise waveforms.

CN118965178AInactive Publication Date: 2024-11-15BEIJING BCIFLEX MEDICAL TECH CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411433500.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-15
Publication Date
2024-11-15
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The existing technology has high-frequency noise interference, low signal-to-noise ratio, unreasonable classification thresholds in neural signal waveform classification, resulting in inconsistent classification results and low accuracy, requiring a large amount of manual labeling to increase costs and time.

Method used

Multiple preset waveform statistical indicators and unsupervised learning target classifier models are used for classification, combined with the preset template library to verify the preliminary classification results and make final classification decisions, reduce manual annotation and judgment time, improve the objectivity of classification results and accuracy.

Benefits of technology

Through automated waveform feature extraction and unsupervised learning models, manual intervention is reduced, classification efficiency and accuracy of action potential and noise waveforms are improved, and inconsistency of classification results is reduced.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118965178A_ABST
    Figure CN118965178A_ABST
Patent Text Reader

Abstract

The invention discloses a neural signal waveform classification method, a computer device and a storage medium, and the method comprises the steps: determining the waveform features of to-be-classified neural signal waveform data according to a plurality of preset waveform statistical indexes, and carrying out the classification of the to-be-classified neural signal waveform data through a target classifier model based on unsupervised learning according to the waveform features, and determining a preliminary classification result of the to-be-classified neural signal waveform data, and determining a final classification result according to the category to which the target template waveform belongs under the condition that the preliminary classification result does not have obvious bias and the target template waveform exists in the preset template library. A plurality of preset waveform statistical indexes are adopted, an unsupervised learning target classifier model is used for classification, the action potential waveform and the noise waveform can be more efficiently distinguished while the classification basis is more objective and reliable, and when the preliminary classification result does not have obvious deviation, the judgment is carried out in combination with a preset template library, so that the judgment accuracy is improved. And the accuracy of the classification result is further improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of biomedical engineering technology, and in particular to a neural signal waveform classification method, a computer device, and a storage medium. Background Art

[0002] When using the spike sorting algorithm to extract and classify neural signals, the output classification results may contain both action potentials from neurons and noise from other sources because high-frequency noise may be mixed in the signal, the signal-to-noise ratio is not high enough, the neuronal action potential waveforms are diverse, the action potential detection threshold is not set reasonably, and the classification algorithm has a weak result filtering ability. Therefore, the neural signal waveforms obtained by the action potential sorting algorithm need to be further classified into "action potential" or "noise".

[0003] At present, when classifying the above-mentioned neural signal waveforms, a large amount of experience-based manual labeling work is often required, which not only increases the manpower and time costs, but also leads to inconsistent classification results due to differences in personnel's judgment methods and levels, thereby reducing the classification accuracy. Summary of the invention

[0004] The embodiments of the present application provide a neural signal waveform classification method, a computer device, and a storage medium. By adopting multiple preset waveform statistical indicators and using an unsupervised learning target classifier model for classification, the efficiency of distinguishing action potential waveforms and noise waveforms is improved.

[0005] In a first aspect, a neural signal waveform classification method is provided, comprising: determining waveform features of neural signal waveform data to be classified according to multiple preset waveform statistical indicators; determining, according to the waveform features, a preliminary classification result of the neural signal waveform data to be classified and a probability value corresponding to the preliminary classification result using a target classifier model based on unsupervised learning, the preliminary classification result including belonging to an action potential or belonging to noise; determining a target waveform that can represent the neural signal waveform data to be classified when the difference between the probability value corresponding to the action potential and the probability value corresponding to the noise is within a preset range; determining whether there is a target template waveform that matches the target waveform based on a correlation between the target waveform and each template waveform in a preset template library, wherein the preset template library includes multiple template waveforms, and the category of each template waveform belongs to action potential or noise; when there is a target template waveform that matches the target waveform in the preset template library, determining the final classification result of the neural signal waveform data to be classified according to the category to which the target template waveform belongs.

[0006] In a second aspect, a computer device is provided, comprising a processor, a memory, and a computer program stored in the memory, wherein the processor executes the computer program to implement the neural signal waveform classification method as described in the first aspect.

[0007] According to a third aspect, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the neural signal waveform classification method according to the first aspect is implemented.

[0008] By applying the above technical solution, the waveform characteristics of the neural signal waveform data to be classified are first determined according to multiple preset waveform statistical indicators, and then the preliminary classification results of the neural signal waveform data to be classified are determined based on the waveform characteristics using the target classifier model based on unsupervised learning. When the preliminary classification results do not have obvious bias and there is a target template waveform in the preset template library, the final classification results are determined according to the category to which the target template waveform belongs. By adopting multiple preset waveform statistical indicators and using the target classifier model of unsupervised learning for classification, the time for manual labeling and discrimination is reduced. While making the classification basis more objective and reliable, the action potential waveform and the noise waveform can be more efficiently discriminated. When the preliminary classification results do not have obvious bias, the preset template library is combined for discrimination, which further improves the accuracy of the classification results. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In the drawings, which are not necessarily drawn to scale, the same reference numerals may describe similar parts in different views. The same reference numerals with letter suffixes or different letter suffixes may represent different instances of similar parts. The accompanying drawings generally illustrate various embodiments by way of example and not limitation, and together with the specification and claims, are used to illustrate the disclosed embodiments. When appropriate, the same reference numerals are used throughout the drawings to refer to the same or similar parts. Such embodiments are illustrative and are not intended to be exhaustive or exclusive embodiments of the present apparatus or method.

[0010] Figure 1 A flowchart showing a neural signal waveform classification method according to an embodiment of the present application is shown; Figure 2 A flowchart of determining a target classification model according to an embodiment of the present application is shown; Figure 3 A flowchart of generating a preset template library according to an embodiment of the present application is shown; Figure 4 A flowchart showing a neural signal waveform classification method according to another embodiment of the present application; Figure 5A schematic diagram showing a neural signal waveform data sample to be classified containing noise and neuron action potential according to an embodiment of the present application; Figure 6 A schematic diagram showing determination of waveform positioning points according to an embodiment of the present application is shown; Figure 7 A schematic diagram showing the principle of a target classifier model based on unsupervised learning according to an embodiment of the present application is shown; Figure 8 A schematic diagram showing multiple template waveforms in a preset template library according to an embodiment of the present application; Fig. 9 A structural block diagram of a computer device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0011] Various aspects and features of the present application are described herein with reference to the accompanying drawings.

[0012] It should be understood that various modifications may be made to the embodiments of the present application. Therefore, the above description should not be considered as limiting, but only as an example of an embodiment. Other modifications within the scope and spirit of the present application will occur to those skilled in the art.

[0013] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present application and, together with the general description of the present application given above and the detailed description of the embodiments given below, serve to explain the principles of the present application.

[0014] These and other characteristics of the present application will become apparent from the following description of a preferred form of embodiment given as a non-limiting example with reference to the accompanying drawings.

[0015] It should also be understood that although the present application has been described with reference to some specific examples, those skilled in the art will be able to readily implement many other equivalent forms of the present application.

[0016] The above and other aspects, features and advantages of the present application will become more apparent in view of the following detailed description when taken in conjunction with the accompanying drawings.

[0017] Specific embodiments of the present application are described hereinafter with reference to the accompanying drawings; however, it should be understood that the embodiments applied for are merely examples of the present application, which may be implemented in a variety of ways. Well-known and / or repeated functions and structures are not described in detail to avoid unnecessary or redundant details that obscure the present application. Therefore, the specific structural and functional details applied for herein are not intended to be limiting, but merely serve as a basis and representative basis for the claims to teach those skilled in the art to use the present application in a variety of ways with substantially any suitable detailed structure.

[0018] This specification may use the phrases "in one embodiment," "in another embodiment," "in yet another embodiment," or "in other embodiments," all of which may refer to one or more of the same or different embodiments according to the present application.

[0019] A neural signal waveform classification method in an embodiment of the present application adopts multiple preset waveform statistical indicators and uses an unsupervised learning target classifier model for classification, thereby reducing the time for manual labeling and discrimination. While making the classification basis more objective and reliable, it can more efficiently discriminate action potential waveforms and noise waveforms. When the preliminary classification results do not have obvious bias, the preset template library is combined for discrimination, thereby further improving the accuracy of the classification results.

[0020] Figure 1 A flowchart of a neural signal waveform classification method according to an embodiment of the present application is shown. Figure 1 As shown, the method comprises the following steps: Step S101, determining waveform features of the neural signal waveform data to be classified according to a plurality of preset waveform statistical indicators.

[0021] In this embodiment, the neural signal waveform data to be classified may be obtained by extracting features and classifying neural signals using an action potential classification algorithm, and the action potential classification algorithm may include any one of the algorithms such as MountainSort and Quicksort. Since high-frequency noise may be mixed in neural signals, the signal-to-noise ratio is not high enough, the action potential waveform of neurons is diverse, the setting of the action potential detection threshold is not reasonable, and the result filtering ability of the classification algorithm is not strong, the neural signal waveform data to be classified may belong to noise. Therefore, it is necessary to further determine whether the neural signal waveform data to be classified belongs to action potential or noise.

[0022] A plurality of waveform statistical indicators are preset, and waveform characteristics of the neural signal waveform data to be classified can be determined according to the plurality of preset waveform statistical indicators.

[0023] Step S102, according to the waveform features, using a target classifier model based on unsupervised learning, determine a preliminary classification result of the neural signal waveform data to be classified and a probability value corresponding to the preliminary classification result, wherein the preliminary classification result includes belonging to an action potential or belonging to noise.

[0024] After obtaining the waveform features, the waveform features are input into the target classifier model of unsupervised learning, and classification is performed using the target classifier model of unsupervised learning to determine the preliminary classification results of the neural signal waveform data to be classified and the probability values ​​corresponding to the preliminary classification results. For example, the preliminary classification result is that it belongs to action potentials, where the probability value of belonging to action potentials is 80%, and the probability value of belonging to noise is 20%.

[0025] Optionally, the target classifier model based on unsupervised learning includes any one of the classification models using algorithms such as self-organizing map, K-means, Gaussian mixture model, DBSCAN (density-based spatial clustering of applications with noise), principal component analysis, t-distribution neighborhood embedding algorithm, autoencoder, singular value decomposition, non-negative matrix factorization, multidimensional scaling, isometric feature mapping, uniform manifold approximation and projection.

[0026] Step S103, when the difference between the probability value corresponding to the action potential and the probability value corresponding to the noise is within a preset range, determining a target waveform that can represent the waveform data of the neural signal to be classified.

[0027] If the difference between the probability value corresponding to the action potential and the probability value corresponding to the noise is within the preset range, it means that the preliminary classification result has no obvious bias. For example, if the preset range is 10%, the probability value corresponding to the action potential is 54%, the probability value corresponding to the noise is 46%, and the difference is 8%. It is determined that the difference is within the preset range. Although the probability of belonging to the action potential is higher than the probability of belonging to the noise, the two probability values ​​are relatively close. In order to improve the accuracy, it is necessary to determine the target waveform that can represent the waveform data of the neural signal to be classified, and then use the target waveform in combination with the preset template library for further discrimination.

[0028] Step S104, based on the correlation between the target waveform and each template waveform in a preset template library, determine whether there is a target template waveform matching the target waveform, wherein the preset template library includes multiple template waveforms, and the category of each template waveform belongs to action potential or noise.

[0029] In this embodiment, a plurality of template waveforms are preset in the template library, each template waveform being an action potential or noise, for example, Figure 8 A schematic diagram showing multiple template waveforms in a preset template library according to an embodiment of the present application is shown, such as Figure 8 As shown, Spike1- Spike12 are template waveforms belonging to action potentials, and Noise1-Noise4 are template waveforms belonging to noises. The correlation between the target waveform and each template waveform can be calculated according to a preset correlation algorithm (such as Euclidean distance, etc.), and whether the target template waveform exists in the preset template library is determined according to each correlation, wherein the correlation between the target waveform and the target template waveform is higher than the preset correlation.

[0030] Step S105 , when there is a target template waveform matching the target waveform in the preset template library, determining a final classification result of the neural signal waveform data to be classified according to the category to which the target template waveform belongs.

[0031] If there is a target template waveform matching the target waveform in the preset template library, it means that there is a template waveform similar to the target waveform in the preset template library, and the category to which the target template waveform belongs is used to determine the final classification result of the neural signal waveform data to be classified. Specifically, if the target template waveform belongs to an action potential, it is determined that the neural signal waveform data to be classified belongs to an action potential. If the target template waveform belongs to noise, it is determined that the neural signal waveform data to be classified belongs to noise.

[0032] The neural signal waveform classification method of the embodiment of the present application first determines the waveform characteristics of the neural signal waveform data to be classified based on multiple preset waveform statistical indicators, and then determines the preliminary classification result of the neural signal waveform data to be classified based on the waveform characteristics using a target classifier model based on unsupervised learning. When the preliminary classification result does not have obvious bias and there is a target template waveform in the preset template library, the final classification result is determined according to the category to which the target template waveform belongs. By adopting multiple preset waveform statistical indicators and using the target classifier model of unsupervised learning for classification, the time for manual labeling and discrimination is reduced. While making the classification basis more objective and reliable, the action potential waveform and the noise waveform can be more efficiently discriminated. When the preliminary classification result does not have obvious bias, the preset template library is combined for discrimination, which further improves the accuracy of the classification result.

[0033] In some implementations of the present application, the neural signal waveform data to be classified includes one or more waveforms, wherein the similarity between the multiple waveforms is higher than a preset similarity, The determining of a target waveform that can represent the neural signal waveform data to be classified specifically includes: In a case where the neural signal waveform data to be classified includes one waveform, determining the one waveform as the target waveform; In the case that the neural signal waveform data to be classified includes multiple waveforms, average values ​​of the multiple waveforms at each time point are determined respectively, and the waveform corresponding to each average value is determined as the target waveform.

[0034] In this embodiment, the neural signal waveform data to be classified includes one or more waveforms. If it includes one waveform, the waveform is directly determined as the target waveform. If the neural signal waveform data to be classified includes multiple waveforms, the multiple waveforms have statistically significant similarities, that is, the similarity between the waveforms is higher than the preset similarity. For example, the multiple waveforms have significant similarities in the characteristics of the resting state, depolarization stage, repolarization stage, hyperpolarization stage, refractory period, and spike amplitude and spike width within a reasonable range. Alternatively, the multiple waveforms do not have the aforementioned characteristics, or only have some of the aforementioned characteristics. The average values ​​of the multiple waveforms in the neural signal waveform data to be classified at each time point are determined respectively, and the waveform corresponding to each average value is determined as the target waveform.

[0035] Optionally, each time point of the waveform may be all corresponding time points on the waveform. It may also be multiple representative time points on the waveform, for example, each time point is a time point corresponding to a zero-crossing point, a peak, a trough, etc. on the waveform. It may also be multiple time points divided at preset time intervals.

[0036] According to the situation that the neural signal waveform data to be classified includes one or more waveforms, the target waveform that can represent the neural signal waveform data to be classified is determined in different ways, thereby ensuring reliable determination of the target waveform.

[0037] In some embodiments of the present application, Figure 2 A flow chart of determining a target classification model according to an embodiment of the present application is shown as follows: Figure 2 As shown, the process of determining the target classifier model includes: Step S201, determining sample waveform features of the neural signal waveform data samples to be classified according to a plurality of the preset waveform statistical indicators.

[0038] In this embodiment, the neural signal waveform data samples to be classified are multiple groups, each group includes multiple waveform samples, for example, Figure 5 A schematic diagram showing a sample of neural signal waveform data to be classified containing noise and neuron action potential according to an embodiment of the present application is shown as follows: Figure 5 As shown, 6 groups of neural signal waveform data samples to be classified are shown, each group of neural signal waveform data samples to be classified includes multiple waveform samples, and each group of neural signal waveform data samples to be classified may belong to action potentials or noise. The sample waveform features of the neural signal waveform data samples to be classified are determined according to multiple preset waveform statistical indicators.

[0039] Step S202: training an initial classifier model based on unsupervised learning according to the sample waveform features.

[0040] The sample waveform features are input into an initial classifier model based on unsupervised learning, and the initial classifier model based on unsupervised learning is trained. The initial classifier model of unsupervised learning and the target classifier model use the same model structure.

[0041] Step S203, when the training completion condition is met, the trained initial classifier model is determined as the target classifier model, and the training completion condition includes reaching a preset number of iterations, or the clustering result no longer changes, or the specified performance indicator no longer improves.

[0042] During the training process of the initial classifier model, it is determined whether the training completion conditions are met. If the training completion conditions are met, the trained initial classifier model is determined as the target classifier model. The training conditions include reaching a preset number of iterations, or the clustering results no longer changing, or the specified performance indicators no longer improving, thereby achieving efficient and accurate determination of the target classifier model.

[0043] For example, Figure 7 A schematic diagram of the principle of the target classifier model based on unsupervised learning according to an embodiment of the present application is shown as follows: Figure 7 As shown, where x i (i = 1,2,3,...,m) is the waveform feature of the input, m is the number of waveform features, corresponding to multiple statistical indicators, w is the weight vector, y is the category to which the output belongs, including y0 and y1, which can represent action potential and noise respectively. If the initial classifier model adopts the self-organizing maps (SOMs) algorithm, the initial classifier model consists of neurons arranged in a grid, each neuron is connected to all input features, and each connection has an associated weight. Therefore, when the dimension of the input data is (N, m) and the dimension of the output data is (N, 2), the dimension of the weight matrix w is (N, m, 2), and each weight in the matrix can be expressed as w ij k , where i=1,2,...,m, j=1,2, k=1,2,...,N.

[0044] When training the initial classifier model, the weights of the neurons can be initialized randomly or using some heuristic methods. An input sample is selected from the neural signal waveform data samples to be classified as the input vector. The Euclidean distance and other distance metrics are used to determine the neuron whose weight vector is closest to the input vector, that is, the best matching unit (BMU). When the Euclidean distance is used to calculate the distance between the weight vector and the input vector, the distance D(j) = Σ(w ij – x i )2 , when the D(j) value is the smallest, the index of j is the index of BMU and can be represented by J.

[0045] After that, the weights of the BMU and its neighboring neurons are iteratively adjusted to make them closer to the input vector. The update formula of the weight vector is: iJ = w iJ (old) + alpha (t) * (x i k -w iJ (old)), where w iJ (old) represents the weight to be updated, and alpha(t) is the learning rate at time t.

[0046] Gaussian function and Mexican hat function can be used to update weights to ensure that neurons close to the BMU have more weight updates than neurons far away. In the iterative process, the learning rate and neighborhood size are gradually reduced as the distance from the BMU and time pass. When the performance of the self-organizing map on the test set is no longer updated for a long time, it is determined that the training completion condition is met and the target classifier model is obtained.

[0047] In some embodiments of the present application, after the trained initial classifier model is determined as the target classifier model, Figure 3 A flow chart of generating a preset template library according to an embodiment of the present application is shown as follows: Figure 3 As shown, the method further comprises the following steps: Step S204, using the target classifier model to respectively determine sample classification results of multiple groups of neural signal waveform data samples to be classified.

[0048] In this embodiment, after obtaining the target classifier model, the target classifier model is used to determine the sample classification results of multiple groups of neural signal waveform data samples to be classified, that is, to determine whether each group of neural signal waveform data samples to be classified belongs to action potentials or noise.

[0049] Step S205 , determining target sample waveforms that can respectively represent each group of neural signal waveform data samples to be classified.

[0050] The average value of each group of neural signal waveform data samples to be classified at each time point can be determined, an average waveform can be formed according to each average value, and each average waveform can be determined as each target sample waveform.

[0051] Step S206: Generate the preset template library using each of the target sample waveforms and the corresponding sample classification results.

[0052] In this embodiment, each target sample waveform and the corresponding sample classification result are saved to form each template waveform, thereby generating a preset template library.

[0053] The target classifier model is used to classify multiple groups of neural signal waveform data samples to be classified respectively, and the classification results and the corresponding target sample waveforms are used to generate a preset template library, thereby realizing the efficient generation of the preset template library. In addition, since the preset template library stores multiple template waveforms and corresponding classification results, it is guaranteed that the target template waveform subsequently determined by the preset template library has good reference.

[0054] In some embodiments of the present application, the preset waveform statistical indicators include at least two of the following: The absolute value of the average value at each time point in the same waveform; The standard deviation of each time point in the same waveform; The standard deviation of the amplitude of each waveform in the same set of waveforms; The absolute maximum value of the waveform; The absolute minimum value of the waveform; The ratio of the absolute maximum value to the absolute minimum value of a waveform; The peak-to-valley difference of the waveform; The valley-to-valley difference of the waveform; Waveform first gradient; Waveform second gradient; Waveform third gradient; The ratio of the sum of the absolute values ​​of the average amplitudes of each waveform in the same group of waveforms to the sum of the standard deviations of the waveforms; The half-width of the peak.

[0055] In this embodiment, the absolute value of the average value at each time point in the same waveform can be expressed as: , where T represents the total duration of the waveform, t represents the time point, Represents the value of the nth waveform at time point t.

[0056] The standard deviation of each time point in the same waveform can be expressed as: ,in, Indicates the average value of the nth waveform at each time point.

[0057] The standard deviation of the amplitude of each waveform in the same set of waveforms can be expressed as: , where N represents the number of waveforms in the same set of waveforms, represents the value of the nth waveform at time point t, Represents the average value of each waveform in the same group of waveforms at each time point.

[0058] The absolute maximum value of the waveform can be expressed as: .

[0059] The absolute minimum value of the waveform can be expressed as: .

[0060] The ratio of the absolute maximum value to the absolute minimum value of a waveform can be expressed as: . The ratio of the sum of the absolute values ​​of the average amplitudes of each waveform in the same group of waveforms to the sum of the standard deviations of the waveforms can be expressed as: .

[0061] The half-peak width of the peak can be expressed as t 2 nAmax / 2 –t 1 nAmax / 2 , where t 2 nAmax / 2 Indicates the value of the nth waveform at half of the falling edge of the peak, t 1 nAmax / 2 Indicates the value of the nth waveform at halfway of the rising edge of the peak.

[0062] It can be understood that the same group of waveforms in the embodiment of the present application refers to the waveforms belonging to the same neural signal waveform data to be classified.

[0063] By using the at least two statistical indicators mentioned above, the waveform characteristics are determined from at least two perspectives, so that the waveform characteristics are more consistent with the actual characteristics of the neural signal waveform data to be classified, and the flexibility is improved.

[0064] In some embodiments of the present application, the peak-to-valley difference of the waveform, the valley-to-valley difference of the waveform, the first gradient of the waveform, the second gradient of the waveform, and the third gradient of the waveform are respectively determined by corresponding waveform positioning points, and the waveform positioning points include: Threshold detection point; The first valley point after the threshold detection point; The first peak point after the threshold detection point; The second zero crossing after the threshold detection point.

[0065] In this embodiment, the threshold detection point may be a detection point determined according to a fixed voltage amplitude, and the fixed voltage amplitude may be, for example, -80 μV to -60 μV. The threshold detection point may also be determined by calculation according to a preset algorithm, for example, -6 to -3.5 times the root mean square of each voltage amplitude in the waveform is determined as the threshold detection point. Figure 6 FIG. 2 is a schematic diagram showing a waveform positioning point determination according to an embodiment of the present application. Figure 6 As shown, the threshold detection point can be represented as P1, the first valley point after the threshold detection point can be represented as P2, the first peak point after the threshold detection point can be represented as P3, and the second zero-crossing point after the threshold detection point can be represented as P4. Correspondingly, the peak-to-valley difference of the waveform is: (AnP3 –A nP2 ). The valley-to-valley difference of the waveform is: (A nP4 –A nP2 ). The first gradient of the waveform is: (A nP2 –A nP1 ) / (t nP2 –t nP1 ). The second gradient of the waveform is: (A nP3 –A nP2 ) / (t nP3 –t nP2 ). The third gradient of the waveform is: (A nP4 –A nP2 ) / (t nP4 –t nP2 ). Among them, A nP1 , A nP2 , A nP3 , A nP4 Respectively represent the amplitude of the nth waveform at P1, P2, P3, and P4, t nP1 ,t nP2 ,t nP3 ,t nP4 They represent the time points P1, P2, P3 and P4 of the nth waveform respectively.

[0066] Corresponding preset waveform statistical indicators are determined through multiple waveform positioning points, thereby improving the accuracy of the preset waveform statistical indicators.

[0067] In some embodiments of the present application, when the target template waveform does not exist in the preset template library, at least one of the following is further included: Determining the preliminary classification result as the final classification result; Send out prompt information for manual classification; The target waveform is manually classified and added to the preset template library as a template waveform.

[0068] In this embodiment, when the target template waveform does not exist in the preset template library, the preliminary classification result can be determined as the final classification result, so as to quickly determine the final classification result. A prompt message for manual classification can also be issued to ensure the accuracy of the final classification result. The target waveform can also be added to the preset template library as a template waveform after manual classification, so as to update the preset template library.

[0069] Optionally, only a prompt message for manual classification may be issued, or after issuing a prompt message for manual classification, the target waveform may be added to a preset template library according to the result of the manual classification.

[0070] In some embodiments of the present application, after determining the preliminary classification result of the neural signal waveform data to be classified and the probability value corresponding to the preliminary classification result according to the waveform features by using the target classifier model based on unsupervised learning, Figure 4 A flowchart of a neural signal waveform classification method according to another embodiment of the present application is shown. Figure 4 As shown, the method further comprises the following steps: Step S106, when the difference between the probability value corresponding to the action potential and the probability value corresponding to the noise is not within the preset range, the preliminary classification result is determined as the final classification result.

[0071] In this embodiment, if the difference between the probability value corresponding to the action potential and the probability value corresponding to the noise is not within the preset range, it means that the preliminary classification result has obvious bias, and the preliminary classification result can be directly determined as the final classification result. For example, if the preset range is 10%, the probability value corresponding to the action potential is 80%, the probability value corresponding to the noise is 20%, and the difference is 60%, then it is determined that the difference is not within the preset range, and the final classification result is determined to be an action potential.

[0072] Based on the same application concept, the embodiment of the present application also provides a computer device, Fig. 9 A structural block diagram of a computer device according to an embodiment of the present application is shown as follows: Fig. 9 As shown, the computer device includes a processor, a memory, and a computer program stored in the memory, and the processor executes the computer program to implement the neural signal waveform classification method as described in various embodiments of the present application.

[0073] The computer device in the embodiment of the present application may be a terminal, or may be other devices other than a terminal. Exemplarily, the computer device may be a mobile phone, a tablet computer, a laptop computer, a PDA, an in-vehicle electronic device, a mobile Internet device (MID), an augmented reality (AR) / virtual reality (VR) device, a robot, a wearable device, an ultra-mobile personal computer (UMPC), a netbook or a personal digital assistant (PDA), etc. It may also be a server, a network attached storage (NAS), a personal computer (PC), a television (TV), a teller machine or a self-service machine, etc., which is not specifically limited in the embodiment of the present disclosure.

[0074] The memory may include a RAM (Random Access Memory) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0075] The above-mentioned processor can be a general-purpose processor, including CPU, NP (Network Processor), etc.; it can also be DSP (Digital Signal Processing), ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0076] In another embodiment provided in the present application, a computer-readable storage medium is also provided, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, the neural signal waveform classification method described in each embodiment of the present application is implemented.

[0077] The computer readable storage medium is such as but not limited to any type of disk including floppy disks, optical disks, CDROMs, magneto-optical disks, read-only memories (ROMs), random access memories (RAMs), EPROMs, EEPROMs, magnetic or optical cards, or any type of medium suitable for storing electronic instructions, each of which is coupled to a computer system bus.

[0078] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integration. The available medium can be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk), etc.

[0079] The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and protection scope of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present application.

Claims

1. A neural signal waveform classification method, characterized in that: include: Determining waveform characteristics of the neural signal waveform data to be classified according to a plurality of preset waveform statistical indicators; According to the waveform features, using a target classifier model based on unsupervised learning, determining a preliminary classification result of the neural signal waveform data to be classified and a probability value corresponding to the preliminary classification result, wherein the preliminary classification result includes belonging to an action potential or belonging to noise; When the difference between the probability value corresponding to the action potential and the probability value corresponding to the noise is within a preset range, determining a target waveform that can represent the waveform data of the neural signal to be classified; Based on the correlation between the target waveform and each template waveform in a preset template library, determining whether there is a target template waveform matching the target waveform, wherein the preset template library includes a plurality of template waveforms, and the category of each template waveform belongs to action potential or noise; In the case that there is a target template waveform matching the target waveform in the preset template library, a final classification result of the neural signal waveform data to be classified is determined according to the category to which the target template waveform belongs.

2. The neural signal waveform classification method according to claim 1, characterized in that: The neural signal waveform data to be classified includes one or more waveforms, wherein the similarity between the multiple waveforms is higher than a preset similarity, The determining of a target waveform that can represent the neural signal waveform data to be classified specifically includes: In a case where the neural signal waveform data to be classified includes one waveform, determining the one waveform as the target waveform; In the case that the neural signal waveform data to be classified includes multiple waveforms, average values ​​of the multiple waveforms at each time point are determined respectively, and the waveform corresponding to each average value is determined as the target waveform.

3. The neural signal waveform classification method according to claim 1, characterized in that: The determination process of the target classifier model includes: Determining sample waveform features of the neural signal waveform data sample to be classified according to the plurality of preset waveform statistical indicators; Training an initial classifier model based on unsupervised learning according to the sample waveform features; When the training completion conditions are met, the trained initial classifier model is determined as the target classifier model. The training completion conditions include reaching a preset number of iterations, or the clustering result no longer changes, or the specified performance indicator no longer improves.

4. The neural signal waveform classification method according to claim 3, characterized in that: After determining the trained initial classifier model as the target classifier model, the method further includes: Determining sample classification results of multiple groups of neural signal waveform data samples to be classified respectively using the target classifier model; Determine target sample waveforms that can respectively represent each group of neural signal waveform data samples to be classified; The preset template library is generated using each of the target sample waveforms and the corresponding sample classification results.

5. The neural signal waveform classification method according to claim 1, characterized in that: The preset waveform statistical indicators include at least two of the following: The absolute value of the average value at each time point in the same waveform; The standard deviation of each time point in the same waveform; The standard deviation of the amplitude of each waveform in the same set of waveforms; The absolute maximum value of the waveform; The absolute minimum value of the waveform; The ratio of the absolute maximum value to the absolute minimum value of a waveform; The peak-to-valley difference of the waveform; The valley-to-valley difference of the waveform; Waveform first gradient; Waveform second gradient; Waveform third gradient; The ratio of the sum of the absolute values ​​of the average amplitudes of each waveform in the same group of waveforms to the sum of the standard deviations of the waveforms; The half-width of the peak.

6. The neural signal waveform classification method according to claim 5, characterized in that: The peak-to-valley difference of the waveform, the valley-to-valley difference of the waveform, the first gradient of the waveform, the second gradient of the waveform, and the third gradient of the waveform are respectively determined by corresponding waveform positioning points, and the waveform positioning points include: Threshold detection point; The first valley point after the threshold detection point; The first peak point after the threshold detection point; The second zero crossing after the threshold detection point.

7. The neural signal waveform classification method according to claim 1, characterized in that: In the case where the target template waveform does not exist in the preset template library, the method further includes at least one of the following: Determining the preliminary classification result as the final classification result; Send out prompt information for manual classification; The target waveform is manually classified and added to the preset template library as a template waveform.

8. The neural signal waveform classification method according to claim 1, characterized in that: After determining the preliminary classification result of the neural signal waveform data to be classified and the probability value corresponding to the preliminary classification result according to the waveform feature by using the target classifier model based on unsupervised learning, the method further includes: When the difference between the probability value corresponding to the action potential and the probability value corresponding to the noise is not within the preset range, the preliminary classification result is determined as the final classification result.

9. A computer device comprising a processor, a memory and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the neural signal waveform classification method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the neural signal waveform classification method according to any one of claims 1 to 8 is implemented.

Citation Information

Patent Citations

  • Defect detection method, electronic equipment and computer readable storage medium

    CN114897806A

  • Neural signal processing method and device, equipment and storage medium

    CN115054266A

  • Joint guard with elastic-based pressure protection

    KR102357960B1