Micro-state analysis method for intrusive cerebral cortex electric signals

Abstract

The invention discloses an intrusive cerebral cortex electric signal micro-state analysis method, which comprises the following analysis steps: step 1, signal acquisition and preprocessing: implanting a cortex surface electrode array into a dog, acquiring a dog ECoG signal, and preprocessing the acquired dog ECoG signal as an original signal; step 2, performing frequency band division processing, and designing a digital filter to decompose the signal into a plurality of frequency bands according to the spectral characteristics of the canine ECoG signal; step 3, segmenting the signal in each frequency band by using a rolling sliding window method, and setting a window length and an overlapping rate; and 4, calculating a signal sample in each sliding window by using a differential entropy index according to a calculation formula which is based on Gaussian distribution hypothesis and directly adopts a non-parametric method 1. The invention provides a new analysis tool for basic neuroscience research, can also be applied to disease diagnosis, behavioral research and intelligent signal processing systems in clinical veterinary medicine, and has remarkable scientific research and application values.

Description

Technical Field

[0001] The present invention relates to the technical field of biomedical signal processing, and more specifically, to a method for microstate analysis of invasive electrocortical signals. Background Art

[0002] With the continuous progress of neuroscience research and signal processing technology, electrocorticogram (ECoG) signals are gradually replacing traditional electroencephalogram (EEG) in clinical neurofunctional localization, epilepsy focus identification, and advanced brain function research due to their high spatial resolution, low artifact interference, and high signal-to-noise ratio. Especially in animal models, ECoG data provides valuable information for in-depth understanding of local activities and dynamic microstates in the cerebral cortex. Currently, existing ECoG microstate analysis methods mainly focus on human signals, and common indicators include power spectral density, phase synchronization, time-domain waveform features, etc. These methods have revealed the stability and dynamic change laws of the brain state to a certain extent. However, for canine electrocortical signals, their signal characteristics are significantly different from those of humans, and there is insufficient capture of local dynamic changes in ECoG data, mainly reflected in the following aspects:

[0003] 1. Signal characteristic differences: Canine electroencephalogram signals have their own characteristics in frequency band distribution, signal-to-noise ratio, and dynamic changes. Directly applying human analysis methods may ignore the unique physiological information of canines.

[0004] 2. Insufficient spatio-temporal resolution: Traditional analysis methods generally use fixed time windows for energy statistics and are difficult to capture short-term and instantaneous state transitions. For the rapid cortical activities in ECoG signals, a higher spatio-temporal resolution analysis method is urgently needed for their microstate changes and non-linear characteristics.

[0005] 3. Limitations in index selection: Traditional methods mainly focus on energy or power spectral density indices and cannot comprehensively reflect the complexity and uncertainty of signals. Differential entropy, as an index for measuring the information content of continuous signals, is more suitable for describing the subtle dynamic changes and complexity in ECoG signals.

[0006] Therefore, there is an urgent need for a new analysis method that is tailored to the characteristics of canine cortical ECoG signals and can accurately capture microstate changes at high spatio-temporal resolution to address the above deficiencies and promote the development of animal neuroscience and clinical veterinary medicine. Summary of the Invention

[0007] Aiming at the problems existing in the prior art, the purpose of the present invention is to provide a method for microstate analysis of invasive electrocortical signals to solve the problems in the background art.

[0008] To achieve the above purpose, the present invention adopts the following technical solutions;

[0009] A microstate analysis method for invasive cerebral cortical electrical signals, comprising the following analysis steps:

[0010] Step 1, signal acquisition and preprocessing: implant a cortical surface electrode array into a canine, collect the canine's ECoG signal, and preprocess the collected canine ECoG signal as the original signal;

[0011] Step 2, frequency band processing: design a digital filter according to the spectral characteristics of the canine ECoG signal to decompose the signal into multiple frequency bands;

[0012] Step 3, segment the signal within each frequency band using a rolling sliding window method, and set the window length and overlap rate;

[0013] Step 4, calculate for each signal sample within the sliding window using the differential entropy index. The calculation formula adopts either the Gaussian distribution hypothesis or the direct non-parametric method, and is solved by statistically analyzing the probability density distribution of the signal within the window;

[0014] Step 5, feature integration and clustering analysis: weight and fuse the differential entropy data of all windows within each frequency band according to the energy ratio to generate a multi-dimensional feature vector sequence;

[0015] Introduce principal component analysis for dimensionality reduction, retain more than 90% of the variance, and reduce redundant information;

[0016] Use the fuzzy C-means clustering (FCM) algorithm to classify the feature vectors, divide the continuous differential entropy sequence into several discrete microstates, and adaptively determine the number of clusters through the gap statistic;

[0017] Step 6, microstate transition and dynamic analysis: according to the clustering results, construct a microstate time series diagram of the canine electroencephalogram signal, analyze the transition rules between different microstates, statistically analyze the duration, transition frequency and transition probability of each microstate, and analyze its correlation with canine behavior, arousal state, and pathological abnormalities;

[0018] Step 7, result visualization and data mining: use visualization methods such as time series diagrams, heat maps, and state transition matrices to intuitively present the microstate analysis results.

[0019] As a further description of the above technical solution:

[0020] The signal acquisition in Step 1 includes recording the canine behavior video and electromyogram signal, and the preprocessing of the original signal includes band-pass filtering, baseline drift correction, and wavelet denoising processing.

[0021] As a further description of the above technical solution:

[0022] In the second step, the canine ECoG signals are divided into δ (0.5–4 Hz), θ (4–8 Hz), α (8–13 Hz), β (13–30 Hz), and γ (30–150 Hz).

[0023] As a further description of the above technical solution:

[0024] In the third step, the rolling window method is to define a window with a fixed length and gradually slide it over the frequency band to divide the continuous signal into multiple segments. The window length and overlap rate are set according to actual requirements.

[0025] As a further description of the above technical solution:

[0026] In the third step, the signals within each time window are regarded as a short-term stationary process, which serves as the basic unit for differential entropy calculation and feature extraction in the fourth step.

[0027] As a further description of the above technical solution:

[0028] In the fourth step, when the data within the window approximately follows a Gaussian distribution, the formula is used:

[0029]

[0030] where σ is the standard deviation of the signal within the window. When the data within the window is non-Gaussian, the differential entropy is directly solved using kernel density estimation.

[0031] As a further description of the above technical solution:

[0032] In the fifth step, each vector contains information of a single frequency band and a description of the comprehensive signal complexity across frequency bands.

[0033] Compared with the prior art, the advantages of the present invention are as follows:

[0034] (1) In this solution, multi-level signal extraction: Through frequency band division processing, the physiological information hidden in each frequency band of the signal can be fully extracted and retained, and the characteristics of canine electroencephalogram signals can be more comprehensively reflected.

[0035] (2) In this solution, high spatio-temporal resolution: By using the rolling window method and high-sampling rate data, the instantaneous dynamic changes of canine electroencephalogram signals are effectively captured, enhancing the spatio-temporal resolution ability of microstate analysis.

[0036] (3) In this solution, innovative complexity quantification: Using differential entropy as an index of signal complexity can better reflect the non-linear, random, and dynamic change characteristics in the signal compared with traditional energy or power spectrum indexes.

[0037] (4) In this solution, data processing with strong adaptability: The clustering algorithm adaptively determines the microstate categories, making the whole method have good generality and robustness, and is applicable to signal analysis under different dog breeds and various behavioral states.

[0038] (5) In this solution, broad application prospects: This method not only provides a new analysis tool for basic neuroscience research, but also can be applied to disease diagnosis (such as epilepsy monitoring), ethology research in clinical veterinary medicine, and intelligent signal processing systems, with significant scientific research and application value. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 is a schematic diagram of the principle of the present invention;

[0040] Figure 2 is a schematic flowchart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0041] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention;

[0042] Please refer to Figure 1 and Figure 2 , the present invention provides the following embodiments:

[0043] A method for microstate analysis of invasive cerebral cortex electrical signals includes the following analysis steps:

[0044] Step 1, signal acquisition and preprocessing: Implant a cortical surface electrode array into the dog, collect the dog's ECoG signal. During the collection process, a high-sampling rate device (such as a sampling rate ≥ 500 Hz) is used to ensure high spatio-temporal resolution data, and the collected dog's ECoG signal is used as the original signal for preprocessing.

[0045] Step 2, frequency band processing: According to the spectral characteristics of the dog's ECoG signal, design a digital filter to decompose the signal into multiple frequency bands. In step 2, the dog's ECoG signal is divided into δ (0.5–4 Hz), θ (4–8 Hz), α (8–13 Hz), β (13–30 Hz), and γ (30–150 Hz). In some experiments, the γ band can be further subdivided to capture high-frequency activities. By adjusting the filter parameters, ensure that the signals in each frequency band have a high signal-to-noise ratio and fully retain the cortical activity information, providing reliable data for subsequent feature calculation.

[0046] Step 3, segment the signals in each frequency band using a rolling sliding window method, set the window length and overlap rate. For example, the window length can be taken as 1–3 seconds, and the overlap rate can be set to 50% or higher to ensure the local stationarity and continuity of the signals within the window.

[0047] Step 4: For the signal samples within each sliding window, calculate using the differential entropy metric. As a measure of the information entropy of continuous signals, differential entropy can reflect the complexity and uncertainty of signals. The calculation formula adopts either the Gaussian distribution assumption or a direct non-parametric method. It is solved by statistically analyzing the probability density distribution of the signals within the window. This step will obtain a series of numerical features representing the signal complexity of each time window.

[0048] Step 5: Feature integration and clustering analysis. Weightedly fuse the differential entropy data of all windows within each frequency band according to the energy ratio to generate a multi-dimensional feature vector sequence;

[0049] Introduce principal component analysis for dimensionality reduction, retaining more than 90% of the variance, reducing redundant information, and reducing the computational amount;

[0050] Use the fuzzy C-means clustering (FCM) algorithm to classify the feature vectors, dividing the continuous differential entropy sequence into several discrete microstates. The number of clusters is adaptively determined by the gap statistic to avoid subjective setting.

[0051] Step 6: Microstate transition and dynamic analysis. According to the clustering results, construct a microstate time series diagram of canine electroencephalogram signals, analyze the transition rules between different microstates, statistically calculate the duration, transition frequency, and transition probability of each microstate, and analyze its correlation with canine behavior, arousal state, and pathological abnormalities (such as epileptic seizures) to provide a basis for subsequent clinical applications or behavior prediction.

[0052] Step 7: Result visualization and data mining. Use visualization methods such as time series diagrams, heat maps, and state transition matrices to intuitively present the microstate analysis results. Combine machine learning algorithms for further pattern recognition and classification prediction of the spatio-temporal distribution of different microstates to achieve real-time monitoring and early warning of canine neural states.

[0053] In the present invention, the signal acquisition in Step 1 includes recording canine behavior videos and electromyogram signals. The preprocessing of the original signals includes band-pass filtering, baseline drift correction, and wavelet denoising processing to remove environmental interference and motion artifacts to obtain high-quality signal data.

[0054] The rolling sliding window method in Step 3 is to define a window with a fixed length and gradually slide it over the frequency band to divide the continuous signal into multiple segments. The window length and overlap rate are set according to actual requirements.

[0055] The rolling sliding window method is a commonly used segmentation technique in time series signal processing, especially suitable for the analysis of ECoG (electrocorticogram) signals. It is applied in frequency band division processing, event detection, and real-time signal processing, and can balance the requirements of time resolution and frequency resolution.

[0056] The signals within each time window in Step 3 are regarded as short-time stationary processes and serve as the basic units for differential entropy calculation and feature extraction in Step 4.

[0057] In Step 4, when the data within the window approximately follows a Gaussian distribution, the formula:

[0058]

[0059] is used, where σ is the standard deviation of the signal within the window. When the data within the window is non-Gaussian, the differential entropy is directly solved using kernel density estimation.

[0060] Each vector in Step 5 contains information of a single frequency band and a description of the comprehensive signal complexity across frequency bands.

[0061] Corresponding data processing and frequency band division methods are designed for the special physiological characteristics of canine ECoG signals, which is conducive to extracting representative frequency band features, so as to apply microstate analysis to the real-time monitoring of canines, provide assistance for canine medical treatment, capture the microstate changes of canine electroencephalogram signals at high spatio-temporal resolution, especially using the rolling sliding window method combined with differential entropy calculation to quantify the signal complexity and uncertainty, improve the accuracy of monitoring, use data mining techniques such as clustering to convert continuous signal indicators into discrete microstates, and further explore the relationship between them and canine behavior or pathological states.

[0062] The above is only a preferred specific embodiment of the present invention; however, the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution of the present invention and its improved concept, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. An invasive microstate analysis method of cerebral cortical electrical signals, characterized in that: The analysis steps include: Step 1: signal acquisition and preprocessing: implanting a cortical surface electrode array into a dog to acquire the dog's ECoG signal, and preprocessing the acquired dog's ECoG signal as an original signal; Step 2: Frequency band processing: According to the spectrum characteristics of canine ECoG signals, a digital filter is designed to decompose the signal into multiple frequency bands; Step 3: Segmentation by rolling window: segment the signal in each frequency band by using rolling window method, and set the window length and overlap rate; Step 4: Differential entropy calculation: for each signal sample in the sliding window, the differential entropy index is used for calculation. The calculation formula adopts one of the Gaussian distribution assumption and the direct non-parametric method, and the probability density distribution of the signal in the statistical window is solved; Step 5: Feature integration and cluster analysis: weighted fusion of the differential entropy data of all windows in each frequency band according to the energy proportion to generate a multi-dimensional feature vector sequence; Introduce principal component analysis to reduce dimension, retain more than 90% of variance and reduce redundant information; The fuzzy C-means clustering (FCM) algorithm is used to classify the feature vectors, and the continuous differential entropy sequence is divided into several discrete micro-states. The number of clusters is adaptively determined by the gap statistics. Step 6: Microstate transition and dynamic analysis: Based on the clustering results, construct a microstate time series diagram of canine EEG signals, analyze the transition rules between different microstates, count the duration, transition frequency and transfer probability of each microstate, and analyze its correlation with canine behavior, wakefulness state and pathological abnormalities; Step 7: Result visualization and data mining: Use the visualization methods of timing diagram, heat map and state transfer matrix to intuitively present the microstate analysis results.

2. The method for invasive microstate analysis of cerebral cortical electrical signals according to claim 1, characterized in that: The signal acquisition in step 1 includes recording the canine behavior video and electromyographic signal, and preprocessing the original signal includes bandpass filtering, baseline drift correction and wavelet noise reduction processing.

3. The method for invasive microstate analysis of cerebral cortical electrical signals according to claim 1, characterized in that: In step 2, the canine ECoG signal is divided into δ (0.5-4 Hz), θ (4-8 Hz), α (8-13 Hz), β (13-30 Hz) and γ (30-150 Hz).

4. The method for invasive microstate analysis of cerebral cortical electrical signals according to claim 1, characterized in that: The rolling window method in step 3 is to define a window of fixed length and slide it step by step on the frequency band to divide the continuous signal into multiple segments. The window length and overlap rate are set according to actual needs.

5. The method for invasive microstate analysis of cerebral cortical electrical signals according to claim 1, characterized in that: The signal in each time window in step three is regarded as a short-term stationary process, which serves as the basic unit for differential entropy calculation and feature extraction in step four.

6. The method for invasive microstate analysis of cerebral cortical electrical signals according to claim 1, characterized in that: In step 4, when the data in the window is close to obeying the Gaussian distribution, the formula is used: Among them, σ is the standard deviation of the signal in the window. When the data in the window is non-Gaussian, kernel density estimation is used to directly solve the differential entropy.

7. The method for invasive microstate analysis of cerebral cortical electrical signals according to claim 1, characterized in that: In step 5, each vector contains information of a single frequency band and a description of the complexity of the integrated signal across frequency bands.

Images