A method and apparatus for classifying neuronal spikes

Abstract

The application relates to a neuron spike sorting method and device, and the method comprises the following steps: acquiring coordinate information of each channel in the high-throughput flexible electrode; determining neighbor channels of a current channel based on the coordinate information of each channel, so as to acquire a spike collected by the current channel and electroencephalogram signals collected by the neighbor channels; acquiring the spike collected by the current channel and the electroencephalogram signals collected by the neighbor channels based on an electroencephalogram signal set to determine a main channel; performing feature extraction and clustering on the electroencephalogram signals collected by the neighbor channels based on the spike to correct the main channel; and acquiring a spike signal collected by the corrected main channel to obtain waveform information of a corresponding neuron. The main channel is determined by dividing a neighborhood based on electrode auxiliary information to obtain a spike signal of a single neuron corresponding to the main channel, the spike sorting accuracy is improved, and stable tracking of spikes of the same neuron is realized.

Description

Technical Field

[0001] This invention relates to the field of electroencephalogram (EEG) signal processing technology, and in particular to a method and apparatus for classifying neuronal spike potentials. Background Technology

[0002] Neuronal spike potential classification is a crucial step in neuronal encoding and decoding. Neuronal encoding and decoding involves converting the electrical signals of neurons into information in some form, such as images or audio, which is then processed and transmitted through the neuronal network. Therefore, accurate spike potential classification is a fundamental requirement for neuroscience research.

[0003] Existing spike potential classification methods mainly have the following technical problems: 1) When using rigid electrodes to collect EEG signals to complete spike potential classification, the rigid electrodes only have a small number of channels in the electrode array, which cannot take advantage of the flexible electrodes that can be implanted for a long time and integrate a large number of electrode channels to record data; 2) In the existing technology, when using flexible electrodes to collect multiple batches of data and classifying each batch of data based on unsupervised classification algorithms, the characteristics of unsupervised classification algorithms result in different neuron numbers each time, making it impossible to achieve stable tracking of spike potentials from the same single neuron.

[0004] Therefore, a classification method is needed that can identify the main channel that collects spike potentials from the same neurons to complete spike potential classification and can stably track spike potentials from the same neurons over a long period of time to solve the above-mentioned technical problems. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a method for classifying neuronal spike potentials. This method solves the problem in existing technologies that, when classifying multiple batches of data collected by flexible electrodes using unsupervised classification algorithms, cannot achieve stable tracking of spike potentials generated by the same single neuron.

[0006] The technical effects of this invention are achieved through the following:

[0007] A method for classifying neuronal spike potentials, comprising:

[0008] Obtain the coordinate information of each channel in the high-throughput flexible electrode;

[0009] Based on the coordinate information of each channel, the neighboring channels of the current channel are determined to obtain the spike potentials collected through the current channel and the EEG signals collected through the neighboring channels.

[0010] The main channel is determined based on the spike potentials acquired by the current channel and the EEG signals acquired by the neighboring channels;

[0011] Based on the spike potential, feature extraction and clustering are performed on the EEG signals collected from the neighboring channels to correct the main channel, resulting in a corrected main channel;

[0012] The modified spike potential signal acquired by the main channel is obtained to obtain the waveform information of the corresponding neuron.

[0013] Further, the coordinate information corresponding to each channel in the high-throughput flexible electrode is obtained, including the following:

[0014] Acquire a set of electroencephalogram (EEG) signals, wherein the data in the set of EEG signals are the EEG signals of neurons collected by each channel of a high-throughput flexible electrode within a continuous time period.

[0015] The data in the set of electroencephalogram (EEG) signals are combined according to their corresponding channels and then filtered to obtain the filtered set of EEG signals.

[0016] Further, determining the neighboring channels of the current channel based on the coordinate information corresponding to each channel includes:

[0017] Obtain the coordinate information of the current channel and the coordinate information of all other channels;

[0018] The distance between any channel and the current channel is obtained based on the coordinate information of the current channel and the coordinate information of all other channels;

[0019] When the distance is less than the preset distance, the current channel is determined to be a neighboring channel of the current channel.

[0020] Furthermore, the continuous time period includes a first time identifier and a second time identifier;

[0021] Both the first time identifier and the second time identifier are within the continuous time period;

[0022] The first time marker is earlier than the second time marker.

[0023] Further, acquiring the spike potential collected by the current channel includes:

[0024] At the first time marker, the first potential acquired by the current channel is obtained;

[0025] At the second time marker, the second potential acquired by the current channel is obtained;

[0026] When the absolute value of the first potential is greater than the potential threshold, and the absolute value of the difference between the first time marker and the second time marker is less than or equal to the minimum allowable time difference, and the first potential is always greater than the second potential, the first potential is the spike potential of the neuron acquired by the current channel.

[0027] Further, acquiring the EEG signals collected by the neighbor channel includes:

[0028] The first time period is obtained by taking a preset interval time before and after the first time marker as the center point;

[0029] The potential of the EEG signal collected by the neighbor channel during the first time period is obtained.

[0030] Further, the main channel is determined based on the spike potentials acquired by the current channel and the EEG signals acquired by the neighboring channels, including:

[0031] The maximum potential value from the potentials of the EEG signals within the first time period is selected and compared with the spike potential;

[0032] When the maximum potential value is less than the spike potential, the current channel is marked as the main channel that generated the spike potential. Neighborhoods are divided using the coordinate information of the channels corresponding to each electrode. This allows for the determination of the main channel and the target neuron corresponding to the spike potential acquired by the main channel by comparing the potentials of the EEG signals acquired by neighboring electrodes with the spike potential of the current neighboring channel. By acquiring the spike potential signals of each main channel, the classification of EEG signal data from multiple neurons in the brain region simultaneously acquired by high-throughput flexible electrodes over a continuous time period is achieved. This effectively improves the accuracy of spike potential signal classification and facilitates stable tracking of spike potential signals from the same neuron.

[0033] Furthermore, feature extraction and clustering of the EEG signals acquired from the neighboring channels based on the spike potential are performed to correct the main channel, including:

[0034] Obtain the potentials of the EEG signals collected by all neighboring channels corresponding to the current channel within the first time period to obtain the target array;

[0035] The eigenvectors corresponding to the target array are determined based on principal component analysis;

[0036] Clustering is performed on the feature vectors to obtain the clustering results;

[0037] The main channels in the clustering results are compared with the main channels determined by the front potential to correct the main channels.

[0038] Further, the main channel in the clustering result is compared with the main channel determined by the front potential to correct the main channel, which includes the following steps:

[0039] The clustering results are matched with the noise data to obtain the corresponding noise overlap.

[0040] When the noise overlap is greater than a preset signal-to-noise ratio, the current clustering result is removed. By extracting segments from the EEG signal data collected from neighboring channels and calculating the overlap with randomly selected noise data, when the overlap exceeds the preset signal-to-noise ratio, the current clustering result is removed, and the clustering results with lower overlap are retained to correct the main channel determined by dividing the neighborhood of the acquired channel's coordinate information, thereby improving the accuracy of the main channel correction.

[0041] Additionally, a classification device for neuronal spike potentials is provided, comprising:

[0042] Acquisition Channel Coordinates Module: Used to acquire the coordinate information of each channel in the high-throughput flexible electrode;

[0043] Neighbor Channel Determination Module: Used to determine the neighbor channels of the current channel based on the coordinate information of each channel, so as to obtain the spike potential collected through the current channel and the EEG signal collected through the neighbor channels;

[0044] Main channel determination module: used to determine the main channel based on the spike potentials acquired by the current channel and the EEG signals acquired by the neighboring channels;

[0045] Main channel correction module: used to perform feature extraction and clustering on the EEG signals collected from the neighboring channels based on the spike potential to correct the main channel and obtain the corrected main channel;

[0046] Spike potential classification module: used to acquire the spike potential signal collected by the corrected main channel to obtain the waveform information of the corresponding neuron.

[0047] As described above, the present invention has the following beneficial effects:

[0048] 1) By dividing the neighborhood using the coordinate information of the acquisition channels corresponding to each electrode, the main channel and the target neuron corresponding to the spike potential of the current neighborhood channel can be determined by comparing the potential of the EEG signal acquired by the neighboring electrodes with the magnitude of the spike potential of the current neighborhood channel. This allows for the classification of EEG signal data of multiple neurons in the brain region acquired simultaneously by multiple channels using high-throughput flexible electrodes within a continuous time period, effectively improving the accuracy of spike potential signal classification and facilitating the stable tracking of spike potential signals of the same neuron.

[0049] 2) By extracting segments from the EEG signal data collected from neighboring channels and calculating the overlap with randomly selected noise data, when the overlap is greater than the preset signal-to-noise ratio, the current clustering result is removed, and the clustering result with lower overlap is retained to correct the main channel determined by dividing the neighborhood of the acquired channel coordinate information, thereby improving the accuracy of correcting the main channel. Attached Figure Description

[0050] To more clearly illustrate the technical solutions of the present invention, the accompanying drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0051] Figure 1 A flowchart illustrating a method for classifying neuronal spike potentials as provided in embodiments of this specification;

[0052] Figure 2 The 32 waveforms are the average values ​​of the EEG signals collected from the 32 channels respectively provided in the embodiments of this specification.

[0053] Figure 3 The waveform of the spike potential acquired by the main channel corresponding to the target neuron provided in the embodiments of this specification;

[0054] Figure 4 This is a block diagram illustrating the composition of a neuronal spike potential classification device provided in the embodiments of this specification. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0057] Example 1:

[0058] like Figure 1 As shown in the embodiments of this specification, a method for classifying neuronal spike potentials is provided, including:

[0059] S100: Obtain the coordinate information of each channel in the high-throughput flexible electrode;

[0060] In this embodiment, high-throughput flexible electrodes are implanted in the intracranial cortex of an organism to collect electroencephalogram (EEG) signals. These flexible electrodes support the simultaneous acquisition of 256 channels of EEG signals from neurons.

[0061] In this embodiment, before performing spike potential classification, all EEG signals are read and preprocessed to obtain a dataset to be classified.

[0062] In one specific implementation, an EEG signal set is acquired, wherein the data in the EEG signal set are the EEG signals of neurons collected by each channel of a high-throughput flexible electrode within a continuous time period; the data in the EEG signal set are combined according to their corresponding channels and filtered to obtain a filtered EEG signal set.

[0063] The EEG signal set was filtered using a bandpass filter with upper and lower bandpass frequencies of 600 Hz and 6000 Hz, respectively.

[0064] S200: Determine the neighboring channels of the current channel based on the coordinate information of each channel, so as to obtain the spike potential collected through the current channel and the EEG signal collected through the neighboring channels;

[0065] In one specific implementation, determining the neighboring channels of the current channel based on the coordinate information corresponding to each channel includes:

[0066] Obtain the coordinate information of the current channel and the coordinate information of all other channels;

[0067] The distance between any channel and the current channel is obtained based on the coordinate information of the current channel and the coordinate information of all other channels;

[0068] When the distance is less than the preset distance, the current channel is determined to be a neighboring channel of the current channel.

[0069] In this embodiment, the coordinate information of the acquisition channel corresponding to the electrode is read, and the following operations are performed on the data on each channel:

[0070] Using the coordinates (x, y) of the current channel as the center, calculate the distance to the current channel for all channels (xi, yi). For all the obtained distances... The system makes a judgment, and if Δs < preset distance, then the channel is considered to be a neighboring channel of the current channel.

[0071] Specifically, the method for obtaining the spike potential acquired through the current channel is as follows:

[0072] The data in the current channel is detected, and the EEG signals corresponding to all time points that meet the following two conditions are marked as a spike potential.

[0073] 1) Select any first time identifier and second time identifier within the continuous time period; wherein both the first time identifier and the second time identifier are within the continuous time period; and the first time identifier is earlier than the second time identifier.

[0074] 2) At the first time marker, acquire the first potential acquired by the current channel; at the second time marker, acquire the second potential acquired by the current channel; when the absolute value of the first potential is greater than the potential threshold, and the absolute value of the difference between the first time marker and the second time marker is less than or equal to the minimum allowable time difference, and the first potential is always greater than the second potential, the first potential is the spike potential of the neuron acquired by the current channel.

[0075] In this embodiment, when both conditions are met simultaneously: |Y(t0)|>threshold and |Y(t0)|>|Y(t1)| are satisfied under the condition that |t0-t1|≤τ, then Y(t0) is a spike potential signal detected by the current channel.

[0076] Where Y(t) represents the EEG signal data acquired by the current channel, and threshold is the potential threshold, calculated as threshold = mean(Y) + μ*std(Y), which is the average of all data Y(t) plus μ times the standard deviation of all data Y(t). τ is the minimum allowed time interval between two detected spike potential signals; that is, within the time interval [t0-τ, t0+τ] centered at t0, it is assumed that at most one spike potential signal exists. Even if multiple Y(t) are detected satisfying the condition Y(t)>threshold, only the Y(t0) with the largest amplitude is retained, and other signals are discarded. μ can be selected by those skilled in the art within the range of 1-10; in this embodiment, μ is taken as 3.5.

[0077] Specifically, the method for acquiring the EEG signals collected through the neighbor channel is as follows:

[0078] A first time period is obtained by taking a preset interval before and after the first time marker as the center point; the potential of the EEG signal collected by the neighbor channel within the first time period is obtained.

[0079] In this embodiment, based on the time point t0 of each spike potential signal detected by the current channel, the time intervals before and after it are respectively the first time intervals τ, to obtain a time interval T with a length of two first time intervals τ, and the potential of the EEG signal collected by the neighboring channel within this time interval is obtained.

[0080] It should be noted that in this embodiment, the high-throughput flexible electrode collects EEG signals at a frequency of 30kHz, and Y(t) is obtained based on this collection frequency. Thus, t is the collection time point corresponding to each collection cycle, that is, Y(t) corresponds to the EEG signal data collected in the corresponding t-th collection cycle.

[0081] T can be any value in the range of 120-180. In this embodiment, T corresponds to 180 time points. Therefore, the entire acquisition time for achieving the classification of potential spikes is 6ms.

[0082] S300: Determine the main channel based on the spike potential acquired by the current channel and the EEG signals acquired by the neighboring channels;

[0083] In one specific implementation, step S300 determines the main channel based on the spike potential acquired by the current channel and the EEG signals acquired by the neighboring channels, including:

[0084] The maximum potential value from the potentials of the EEG signals within the first time period is selected and compared with the spike potential;

[0085] When the maximum potential value is less than the spike potential, the current channel is marked as the main channel that generates the spike potential.

[0086] Specifically, since a neuron typically emits the strongest signal in the main channel (the single channel in which the spike potential of a single neuron is detected), while the peak signals in adjacent channels are lower, the channel corresponding to the maximum potential within a specific time period in the current channel and its corresponding neighborhood is selected as the main channel for detecting the spike potential of a single neuron. The specific determination process is as follows:

[0087] Examine the potential data in each neighboring channel within this time period, find the maximum potential value, and if the maximum potential value is less than or equal to the peak potential signal Y(t0) in the current channel, then temporarily mark the current channel as the main channel corresponding to the peak potential signal.

[0088] If the maximum potential value is greater than the spike potential signal Y(t0) in the current channel, then the neighboring channel corresponding to the maximum potential value is temporarily marked as the main channel.

[0089] S400: Based on the spike potential, feature extraction and clustering are performed on the EEG signals collected from the neighboring channels to correct the main channel, thereby obtaining the corrected main channel;

[0090] In one specific implementation, step S400 involves extracting features and clustering the EEG signals acquired from the neighboring channels based on the spike potentials to correct the main channel, including:

[0091] S410: Obtain the potentials of the EEG signals collected by all neighboring channels corresponding to the current channel within the first time period, and obtain the target array;

[0092] S420: Determine the eigenvectors corresponding to the target array based on principal component analysis;

[0093] In this embodiment, for each time point t0 where a spike potential signal is detected, an array of size M*T is extracted from the data, where M is the number of neighboring channels, T is the size of the data extracted with t0 as the center, and T = 2τ.

[0094] Suppose that a spike event is detected at time t0 on a certain channel. Then, taking time intervals τ before and after t0, we obtain a data set of length T. Next, we search for all neighboring channels and take data of length T at the same time points. Assuming there are M neighboring channels, we obtain an array of size M*T.

[0095] For example, if a spike event is detected at t0 in channel 1, and the value of T is 5, and there are 2 neighboring channels, the data obtained in channel 1 is [1,2,3,4,5]. Then, data of size T are taken from the two neighboring channels with t0 as the center position. Combining them together gives an M*T array, specifically including: [[1,2,3,4,5],[1,1,1,1,1],[0,0,0,0,0]]. This example reduces the amount of data and the number of channels. In actual applications, M is usually 5-10, and T is usually 120 or 180.

[0096] For each detected time point t0, we assume that there is a spike potential signal at that point, and extract an array of size M*T for each t0.

[0097] When L spike potential signals are detected, an array of size L*M*T is obtained. Principal component analysis is performed on this array, and the first n principal components are used as spike potential feature coefficients. n is usually taken as 10 to achieve a dimensionality reduction effect. The generated feature vector array is of size L*n, where L is the number of detected spike potential events and n is the number of retained principal components.

[0098] S430: Cluster the feature vectors to obtain clustering results;

[0099] Specifically, due to the relatively low dimensionality of the feature space, the initial results may contain duplicates. Therefore, the extracted feature data is clustered using the ISO-SPLIT algorithm to obtain preliminary clustering results, namely, the label of each neuron and the main channel corresponding to that label. For clustering results where the number of identical labels exceeds the algorithm's set parameter `min_size`, clustering is reapplied, and this process is repeated until no further cluster splits occur.

[0100] The specific clustering process is as follows:

[0101] 1. Based on the feature vector array L*n, obtain L initial clusters containing n principal components, specifically S1, S2, ..., and assign an initial label to each initial cluster.

[0102] 2. Repeat the following operations:

[0103] a. Calculate the projection direction between any two selected clusters S1 and S2, and project the data points in each cluster onto this direction to obtain X1 and X2. Specifically, as follows:

[0104] Calculate the mean vectors u1 and u2 for all data points in S1 and S2, representing the center positions of S1 and S2, respectively. Calculate the covariance matrices Sigma1 and Sigma2 for all data points in S1 and S2.

[0105] Calculate the projection direction V between S1 and S2 such that the variance between the two clusters is maximized under this direction. Specifically, the projection direction can be calculated as follows:

[0106] 1) Calculate the total covariance matrix of S1 and S2: Sigma = Sigma1 + Sigma2.

[0107] 2) Calculate the eigenvectors and eigenvalues ​​of Sigma, sort the eigenvalues ​​from largest to smallest, and select the eigenvector corresponding to the largest eigenvalue as the projection direction V.

[0108] 3) Project all data points in S1 and S2 along the V direction to obtain X1 and X2. Specifically, for data point yj, its projection in the projection direction V is: Project(V,yj)=(yj-u)*V. Where u=(u1+u2) / 2 represents the center position of S1 and S2, and * represents the vector dot product operation.

[0109] 4) Return the projected datasets X1 and X2 to calculate the optimal split point.

[0110] b. Calculate the optimal split point c between X1 and X2 to determine whether the cluster needs to be further divided into two sub-clusters. If so, reassign the cluster labels based on the split point c. The calculation process for the optimal split point is as follows:

[0111] 1) Arrange the data points in X1 and X2 in ascending order;

[0112] 2) Starting from the first data point in X1, use it as the split point to divide the data points in X1 and X2 into two new subsets respectively, and calculate their mean difference;

[0113] 3) Move the split point to the next data point in X1, repeat step 2 above, and continue to move the split point to the right until the last data point of X1 is reached;

[0114] 4) In step 3 above, record the position c of the split point that maximizes the mean difference and its corresponding mean difference.

[0115] 5) Using c as the optimal split point, divide the data points in X1 and X2 into two new subsets respectively.

[0116] If the mean difference is negative, it means that the two clusters cannot be separated along that projection direction, and in this case, the two clusters should be merged. If the mean difference is zero, it means that the split point can be chosen arbitrarily. In this case, any position can be chosen as the optimal split point.

[0117] c. If no further clustering is required, merge the selected clusters into one cluster and reassign labels.

[0118] d. Mark the cluster pairs that have been used and continue to the next loop.

[0119] 3. Renumber the cluster labels using natural numbers starting from 1.

[0120] 4. Return the cluster labels.

[0121] In the returned clustering results, the number of results with the same label is counted. Due to the large amount of data, there may be cases where a single clustering may not yield accurate results. Therefore, a parameter min_size is set, which is the minimum number of results with the same label contained in a single cluster.

[0122] Clusters with fewer than this value are left unprocessed, while clusters with more than this value are considered to be potentially underclustered. The ISO-SPLIT algorithm is then used again for clustering until no new results are generated.

[0123] If k labels are generated based on the clustering results, then it is considered that k neurons were detected. Figure 2 As shown, Figure 2 This represents the waveform of a neuron after it has been captured by a 32-channel electrode.

[0124] Figure 3 The 32 waveforms are calculated by taking the average value of the EEG signal from 180 points centered on the time point corresponding to the spike potential signal of each labeled neuron. The horizontal axis of the waveform represents the 180 points, and the vertical axis represents the voltage amplitude in µV.

[0125] like Figure 3 As shown, the spike potential classification method of this application can be used to obtain the spike potential waveform corresponding to the main channel of the neuron, i.e. Figure 3 for Figure 2 An enlarged view of the bottom waveform in the third column from left to right.

[0126] Figure 2 Except Figure 3The other waveforms are obtained by averaging the EEG signals in the corresponding channels for the same time period based on the spike potential signals collected from the main channel in the other 31 channels.

[0127] S440: Compare the main channel in the clustering result with the main channel determined by the front potential to correct the main channel.

[0128] Since spike potential signal detection is applied to a single channel, and the flexible electrode multi-channel detection method may result in the same neuron being identified multiple times, it is necessary to merge the results and designate a single main channel.

[0129] Specifically, the clustering results obtained in step 430 are processed as follows:

[0130] For each channel and its neighboring channels, calculate the time points at which a single channel is clustered into the same label. Using this time point as the center, take a data point of size `clip_size` as the waveform, calculate the average, and obtain an M*C*K template, where M is the number of neighboring channels, C is `clip_size`, and K is the number of clustering results. Here, `clip_size` can be equivalent to `T` in step S300.

[0131] In the template, for the waveform mean of K clustering results on different channels, when it meets the following condition: At that time, retain this waveform and consider the main channel of the result to be m.

[0132] Among them, W m,i (m,t) represents the waveform of length t obtained from the i-th clustering result on the m-th channel. The algorithm parameters can be any value within the range of 0-1. In this embodiment... It is 0.9.

[0133] The result of the above process, main channel m, is compared with the main channel marked in step S300. If they are different, the result in step S300 is corrected.

[0134] The purpose of the clustering process is to label each spike potential signal detected at time point t0 with a corresponding label k. Each label is equivalent to labeling a neuron, and the goal is to assign a unique main channel to each neuron corresponding to each label.

[0135] For each label k, find the time point t0 of all spike potential signals belonging to that label, and take the data of length clip_size as the center of the time point t0 as the waveform signal of the neuron.

[0136] These waveform data have the same length and the same label, and may come from the same or different channels. Then, the average value of all waveform signals belonging to the same channel is calculated.

[0137] For example, there are 10 time points for label 1, each corresponding to a different time point. Taking each time point as the center, 120 points are taken as waveforms. Of these 10 waveforms, 2 come from the first channel, 4 from the second channel, and 4 from the third channel. The average of the 120 points from the same channel is then calculated to obtain the 120 points from the first channel, the 120 points from the second channel, and the 120 points from the third channel. These represent the average values ​​of the EEG signals of the neurons corresponding to label 1 collected on these three channels.

[0138] Then, when At that time, the maximum value of a single point in the average value of each channel is calculated and compared. For example, the maximum value of 120 points in the first channel is 70, the second channel is 60, and the third channel is 30. Since 70 > 0.9 * 60 > 0.9 * 30, we consider the first channel to be the main channel of the neuron corresponding to label 1. Then, we compare this result with the main channel result obtained in step S300 to complete the correction.

[0139] In one specific implementation, step S440, which compares the main channel in the clustering result with the main channel determined by the spike potential to correct the main channel, includes the following steps beforehand:

[0140] The clustering results are matched with the noise data to obtain the corresponding noise overlap.

[0141] When the noise overlap is greater than the preset signal-to-noise ratio, the current clustering result is removed.

[0142] For each clustering result, the degree of overlap between each clustering result and the randomly selected noise is calculated. The specific process is as follows:

[0143] For each label obtained from clustering, representing a neuron, waveform data is extracted based on its occurrence time, following the same principle as feature extraction in step S420. Based on the previous clustering results, k neurons corresponding to k labels can be determined, and the number Q of spike potential signals corresponding to each of the k neurons can be obtained. That is, L spike potential signals are clustered into k cluster arrays containing Q spike potential signals, and the k cluster arrays are represented as A1, B1, ...

[0144] Extract a Q*M*T array A1. At the same time, randomly extract an array B1 from the original data with the same number of data points as array A1. Consider B1 as noise data.

[0145] Principal component analysis is performed on both based on step S420, and the first n principal components are retained to obtain two eigenvectors A2 and B2 of size L*n.

[0146] Each row of data in A2 represents an extracted feature, signifying the occurrence of a spike potential. Two new arrays, L1 and L2, are generated. L1 represents the flag for each spike potential in A2, set to all values ​​of 1. L2 represents the flag for each row of data in B2, set to all values ​​of 2. The following formula is used for calculation:

[0147]

[0148] Assume the A2 array after principal component analysis is as follows:

[0149] [[1,1,1],[2,2,2],[3,3,3],[4,4,4],[5,5,5],[6,6,6]],

[0150] Array B2 is:

[0151] [[1,2,3],[3,2,1],[4,5,6],[7,8,9],[9,8,7],[11,12,13]],

[0152] The label array L1 is then [1,1,1,1,1,1], and L2 is [2,2,2,2,2,2]. For each row of elements in A2 and B2, the k-nearest neighbor algorithm is used to obtain the k nearest neighbors of each row element. Then, the label corresponding to these k elements is obtained from the label array.

[0153] If k is 6, taking the first row element [1,1,1] in A2 as an example, assume that its nearest elements are calculated as follows:

[0154] [2,2,2],[3,3,3],[4,4,4],[1,2,3],[5,5,5],[6,6,6], we can obtain an array of length k for the first row of elements in A2 as [1,1,1,2,1,1].

[0155] Performing the above process on each row of elements in A2 and B2 will result in an array of (2*L)*k, which represents the k nearest neighbors of each element.

[0156] Based on the calculation m noiseFormula (A) iterates through each element to calculate the number of elements whose k nearest neighbors have the same label as the element itself. Taking the first row of A2 as an example, the element itself is labeled as 1, and the k nearest neighbor array is [1,1,1,2,1,1]. It can be seen that the number of elements with the label 1 in this array is 5, that is, the number of elements with the same label [1,1,1] in the first row of A2 is 5.

[0157] Count the number of times this label is the same for all elements in A2 and B2, sum them up, and then divide the sum by (2*L*k) to get m in the formula. noise (A), using 1-m noise (A) Compare with the preset overlap parameter noise_threshold, when 1-m noise When (A) > noise_threshold, it is considered that the overlap between the neuron and the noise is too high, and the result of the neuron is discarded. In this embodiment, noise_threshold is set to 0.15.

[0158] In some other implementations, the results can be filtered by the following method before filtering by noise overlap:

[0159] If a spike potential is clustered into the same label, and the time interval between two different events is less than the set parameter detect_interval, then these times and labels are removed from the results.

[0160] This utilizes knowledge from neuroelectrophysiology. There is a minimum time interval, detect_interval, between two adjacent spikes emitted by a neuron. If the interval between two spikes is less than this minimum time interval, the signal is considered not to be a spike signal.

[0161] S500: Acquire the spike potential signal collected by the corrected main channel to obtain the waveform information of the corresponding neuron.

[0162] Based on the coordinate information of the main channel, i.e. the spatial distribution of the main channel, the waveform information of a single neuron is drawn. Each waveform is the average value of all action potentials generated by the same neuron, and the relevant information of the neuron is marked.

[0163] like Figure 4 As shown in the embodiments of this specification, a classification device for neuronal spike potentials is provided, comprising:

[0164] Acquisition Channel Coordinate Acquisition Module 401: Used to acquire the coordinate information of each channel in the high-throughput flexible electrode;

[0165] Neighbor channel determination module 402: used to determine the neighbor channels of the current channel based on the coordinate information of each channel, so as to obtain the spike potential collected through the current channel and the EEG signal collected through the neighbor channels;

[0166] Main channel determination module 403: used to determine the main channel based on the spike potential acquired by the current channel and the EEG signals acquired by the neighboring channels;

[0167] Main channel correction module 404: used to perform feature extraction and clustering on the EEG signals collected from the neighboring channels based on the spike potential to correct the main channel and obtain the corrected main channel;

[0168] Spike potential classification module 405: used to acquire the spike potential signal collected by the corrected main channel to obtain the waveform information of the corresponding neuron.

[0169] While the present invention has been described through preferred embodiments, it is not limited to the embodiments described herein, and various changes and modifications are made without departing from the scope of the invention.

[0170] Where there is no conflict, the above embodiments and features described herein can be combined with each other.

[0171] The above description is merely a preferred embodiment of the present invention and should not be construed as limiting the scope of the invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A method of classifying neuronal spiking, characterized by, The method comprises the following steps: obtaining coordinate information of each channel in a high-throughput flexible electrode; determining neighbor channels of a current channel based on the coordinate information of each channel to obtain a spike collected by the current channel and an electroencephalogram signal collected by the neighbor channel; determining a main channel according to the spike collected by the current channel and the electroencephalogram signal collected by the neighbor channel, the main channel being used to represent a channel corresponding to a maximum potential value of the current channel and its corresponding neighbor channel in a specific time period; obtaining potentials of the electroencephalogram signals collected by all the neighbor channels corresponding to the current channel in a first time period to obtain a target array; determining a feature vector corresponding to the target array based on principal component analysis; performing clustering on the feature vector to obtain a clustering result; comparing the main channel in the clustering result with the main channel determined by the spike to correct the main channel, and obtaining a corrected main channel; obtaining a spike signal collected by the corrected main channel to obtain waveform information of a corresponding neuron.

2. The method of classifying neuronal spiking according to claim 1, wherein, The method for obtaining the coordinate information of each channel in the high-throughput flexible electrode comprises the following steps: obtaining an electroencephalogram signal set, data in the electroencephalogram signal set being electroencephalogram signals of neurons collected by each channel of a high-throughput flexible electrode in a continuous time period; 3. The method of classifying neuronal spiking according to claim 2, wherein, merging and filtering the data in the electroencephalogram signal set according to the channels corresponding to the data to obtain a filtered electroencephalogram signal set. The method for determining the neighbor channels of the current channel based on the coordinate information of each channel comprises the following steps: obtaining coordinate information of the current channel and coordinate information of all other channels; 4. The method of classifying neuronal spike according to claim 3, wherein, obtaining a distance between any channel and the current channel according to the coordinate information of the current channel and the coordinate information of all other channels; when the distance is less than a preset distance, determining that the current channel is a neighbor channel of the current channel. The continuous time period comprises a first time identifier and a second time identifier; 5. The method of classifying neuronal spike according to claim 4, wherein, the first time identifier and the second time identifier are both in the continuous time period; the first time identifier is earlier than the second time identifier. The method for obtaining the spike collected by the current channel comprises the following steps:

6. The method of classifying neuronal spike according to claim 5, wherein, obtaining a first potential collected by the current channel at the first time identifier; obtaining a second potential collected by the current channel at the second time identifier; 7. The method of classifying neuronal spike according to claim 6, wherein, when an absolute value of the first potential is greater than a potential threshold value, and an absolute value of a difference between the first time identifier and the second time identifier is less than or equal to a minimum allowable time difference, and the first potential is always greater than the second potential, the first potential is a spike of a neuron collected by the current channel. The method for obtaining the electroencephalogram signal collected by the neighbor channel comprises the following steps: obtaining a first time period based on the first time identifier as a center point and taking a preset interval time before and after the center point; obtaining potentials of the electroencephalogram signals collected by the neighbor channel in the first time period. The method for determining the main channel according to the spike collected by the current channel and the electroencephalogram signal collected by the neighbor channel comprises the following steps: selecting a maximum potential value from the potentials of the electroencephalogram signals in the first time period and comparing the maximum potential value with the spike. When the maximum potential value is less than the sharp potential, the current channel is marked as a main channel generating the sharp potential.

8. The method of classifying neuronal spike according to claim 7, wherein, The comparison of the main channel in the clustering result with the main channel determined through the sharp potential for correction of the main channel includes: matching the clustering result with noise data to obtain a corresponding noise overlap degree. When the noise overlap degree is greater than a preset signal-to-noise ratio, the current clustering result is removed.

9. An apparatus for classifying neuronal spikes, characterized by The method comprises the steps of: A channel coordinate acquisition module is configured to acquire coordinate information of each channel in a high-throughput flexible electrode. A neighbor channel determination module is configured to determine neighbor channels of a current channel based on the coordinate information of each channel, to acquire a sharp potential collected through the current channel and an electroencephalogram signal collected through the neighbor channels. A main channel determination module is configured to determine a main channel according to the sharp potential collected through the current channel and the electroencephalogram signal collected through the neighbor channels, the main channel being used to represent a channel corresponding to a maximum potential value of the current channel and the corresponding neighbor channels in a specific time period. A target data determination module is configured to acquire potentials of electroencephalogram signals collected by all neighbor channels corresponding to the current channel in a first time period, to obtain a target array. A feature vector determination module is configured to determine a feature vector corresponding to the target array based on principal component analysis. A clustering module is configured to cluster the feature vector to obtain a clustering result. A main channel correction module is configured to compare the main channel in the clustering result with the main channel determined through the sharp potential to correct the main channel, to obtain a corrected main channel. A sharp potential classification module is configured to acquire a sharp potential signal collected by the corrected main channel, to obtain waveform information of a corresponding neuron.

Images