A pathological high frequency oscillation recognition system based on SEEG

Abstract

The application provides a pathological high-frequency oscillation identification system based on SEEG, a signal acquisition module acquires stereoelectroencephalogram data of a target object; a node preliminary detection module performs preliminary detection on M channel signals after preprocessing, and determines suspicious nodes meeting the conditions; a scoring module performs cross-channel time-frequency coherence analysis and cross-channel propagation consistency analysis on each suspicious node, and determines cross-channel propagation consistency scores; and a decision module determines whether pathological high-frequency oscillation exists in the stereoelectroencephalogram data based on the cross-channel propagation consistency scores. The scheme designs a progressive architecture of preliminary detection + cross-channel time-frequency coherence analysis + cross-channel propagation consistency analysis, the preliminary detection has high running efficiency and can quickly detect suspicious nodes, while the cross-channel time-frequency coherence analysis and the cross-channel propagation consistency analysis can perform in-depth analysis, effectively distinguish other similar abnormal discharge signals, and realize high-precision detection of pathological high-frequency oscillation.

Description

Technical Field

[0001] The present application relates to the field of EEG abnormality detection, and in particular to a SEEG-based pathological high-frequency oscillation recognition system. Background Art

[0002] Stereoelectroencephalography (SEEG or stereo EEG) is a minimally invasive technique that precisely locates the starting area of ​​epileptic seizures by implanting electrodes directly into the brain. This method is mainly used for the location of epileptic lesions and the evaluation of brain functional areas, and is characterized by high precision and low trauma. Stereotactic EEG combines stereotactic technology and electroencephalography to record brain electrical activity by implanting electrodes at specific locations in the skull. These electrodes are usually inserted through small holes in the skull and placed in different parts of the brain to cover areas related to epileptic seizures or brain areas related to specific functions. In this way, high-resolution EEG signals can be obtained to help doctors determine the origin and propagation pattern of epileptic seizures, as well as identify brain areas associated with specific cognitive or motor functions.

[0003] SEEG can directly capture abnormal electrical activities that cannot be detected by scalp EEG, such as high-frequency oscillations (fast EEG oscillations of 80~500Hz, such as Ripples, Fast Ripples), sharp waves / spike waves (focal, high-amplitude sharp waveforms, lasting <70ms), low-frequency repetitive discharges (rhythmic spike-slow complexes of 2~10Hz), etc.

[0004] SEEG provides the data foundation for abnormal EEG detection, but the accurate identification of pathological high-frequency oscillations remains a key challenge at this stage. A single person's SEEG EEG signal data volume can reach 40Gb per day. However, for those with pathological high-frequency oscillations, the total duration of these oscillations typically does not exceed 30 seconds, with no clear temporal pattern. Furthermore, abnormal discharge signals with similar characteristics to pathological high-frequency oscillations (such as physiological high-frequency oscillations, electromyographic artifacts, and equipment noise) are also challenging to detect. Traditional approaches to detecting pathological high-frequency oscillations suffer from high false positive rates and low operational efficiency. Therefore, developing a system more suitable for detecting pathological high-frequency oscillations in SEEG is a technical challenge that needs to be addressed in this field. Summary of the Invention

[0005] The purpose of the embodiments of the present application is to provide a pathological high-frequency oscillation identification system based on SEEG, providing a systematic and effective means for efficiently and accurately detecting pathological high-frequency oscillations in SEEG.

[0006] In order to achieve the above objectives, the embodiments of the present application are implemented in the following manner:

[0007] The embodiment of the application provides a pathological high-frequency oscillation identification system based on SEEG, which comprises:

[0008] A signal acquisition module is configured to acquire stereoelectroencephalogram (SEEG) data of a target object, wherein the SEEG data comprises M channel signals.

[0009] A node preliminary inspection module is configured to perform preliminary inspection on the M channel signals after preprocessing, and determine suspicious nodes meeting a condition.

[0010] A scoring module is configured to perform cross-channel time-frequency coherence analysis and cross-channel propagation consistency analysis on each suspicious node, and determine a cross-channel propagation consistency score.

[0011] A decision module is configured to determine whether pathological high-frequency oscillation exists in the SEEG data based on the cross-channel propagation consistency score.

[0012] Further, the node preliminary inspection module is specifically configured to:

[0013] The M channel signals are subjected to band-pass filtering and power frequency notch filtering, and the 80-500Hz frequency band is retained and the 50Hz power frequency interference is removed.

[0014] For each channel signal after preprocessing:

[0015] A sliding window of a first size is used to perform dynamic sliding window analysis on the channel signal, the channel energy of each sliding window is calculated, and a target sliding window with channel energy exceeding a dynamic energy threshold is determined, and the starting position of the target sliding window is determined as a suspicious node.

[0016] Further, the node preliminary inspection module is specifically configured to:

[0017] The channel energy of each sliding window is calculated in the following manner:

[0018]

[0019] wherein, represents the high-frequency energy index of the channel at the end of the th sliding window, , is the starting time of the th sliding window, , the unit , is the initial time, is the end time of the th sliding window, is the first size, , is the signal frequency, is the channel The channel signal after preprocessing, is the short-time Fourier transform;

[0020] like , determine the channel Middle The sliding window is the target sliding window, and the channel is determined Middle The start time of the sliding window is a suspicious node, where satisfy:

[0021] ,

[0022] ,

[0023] ,

[0024] in, is the dynamic energy threshold, is the baseline energy value, express The median absolute deviation of .

[0025] Furthermore, the scoring module is specifically used to:

[0026] For each suspicious node: determining an expansion window of a second size with the suspicious node as a starting position, wherein the second size is larger than the first size, and the expansion window covers each channel;

[0027] For each extended window: perform cross-channel time-frequency coherence analysis on each channel signal in the extended window to determine the coherence characteristics corresponding to the extended window; based on the coherence characteristics corresponding to the extended window, perform cross-channel propagation consistency analysis on the channel signals in the extended window to determine the cross-channel propagation consistency score.

[0028] Furthermore, the scoring module is specifically used to:

[0029] Perform time-frequency transformation on each channel signal in the expansion window to determine the time-frequency spectrum corresponding to each channel;

[0030] Based on the time-frequency spectrum corresponding to each channel, the dynamic coherence between every two channels is calculated;

[0031] The dynamic coherence between every two channels is processed to determine the coherence characteristics corresponding to the expanded window.

[0032] Furthermore, the scoring module is specifically used to:

[0033] For each channel signal in the expansion window, the time-frequency spectrum corresponding to the channel is calculated in the following way:

[0034] ,

[0035] in, Indicates channel The corresponding time-frequency spectrum is For time point, , For the second size, For the The start time of the sliding window, that is, the start time of the extended window, is the signal frequency, , For channel The channel signal after preprocessing, is the integration variable, is the Morlet wavelet, is the scale parameter, represents the complex conjugate.

[0036] Furthermore, the scoring module is specifically used to:

[0037] For every two channels, the dynamic coherence between the two channels is calculated as follows:

[0038] ,

[0039] in, Indicates channel With channel Between the time points and frequency Dynamic coherence on , is the time index, , 、 is the wavelet coefficient, is the complex conjugate of the wavelet coefficient.

[0040] Furthermore, the scoring module is specifically used to:

[0041] For the dynamic coherence between every two channels, adaptive processing is performed in the following way:

[0042] ,

[0043] in, The channel after adaptive processing With channel Between the time points and frequency The dynamic coherence on Channels in the extended window With channel The mean dynamic coherence between For channel With channel The standard deviation of the dynamic coherence between

[0044] The dynamic coherence between each two channels after adaptive processing at each time point in the integrated expansion window , and get the coherence characteristics corresponding to the expanded window:

[0045] ,

[0046] in, Indicates channel At the time point and frequency The coherence eigenvalue when Indicates that the channel All channels except Take and Channel The maximum value of the dynamic coherence, For each channel Perform operations to generate a three-dimensional data structure containing all channels. is the length of the frequency dimension, Represents a three-dimensional tensor.

[0047] Furthermore, the scoring module is specifically used to:

[0048] For each time point in the extended window and frequency , if the channel The coherence eigenvalue of Exceeding the threshold ,mark is an active node;

[0049] For any channel and channel If there is more than the set ratio make and At the same time, it is an active node, and the channel and channel Classify into the same associated channel group, and finally generate an associated channel group , , is the total number of associated channel groups;

[0050] For each associated channel group: calculate the propagation path integral of each channel in the associated channel group, and determine the cross-channel propagation consistency score based on the propagation path integral.

[0051] Furthermore, the scoring module is specifically used to:

[0052] The channels in the associated channel group are calculated in the following way With channel Phase difference:

[0053] ,

[0054] ,

[0055] in, Associate channel group Channels in With channel At the time point The instantaneous phase difference, For time point, , For the second size, For the The start time of the sliding window, that is, the start time of the extended window, For channel The analytical signal, For channel The complex conjugate of the analytical signal, represents the argument of a complex number, For channel The channel signal after preprocessing, is an imaginary number, is the Hilbert transform;

[0056] The associated channel group is calculated in the following way Channels in With channel Propagation parameters:

[0057] ,

[0058] in, Indicates the associated channel group Channels in Towards channel The propagation intensity of directional propagation, Instantaneous phase difference The time gradient of is a symbolic function that reveals the channel With channel The direction of propagation between

[0059] The associated channel group is calculated in the following way Channels in The propagation path integral of :

[0060] ,

[0061] in, Indicates the associated channel group Channels in The propagation path integral reveals the channel The net transmission intensity, Associate channel group The number of channels in is a half-wave rectifier function, retaining only positive values, Indicates the associated channel group Channels in The total outflow intensity, Indicates the associated channel group Channels in Total inflow intensity;

[0062] The associated channel group is calculated in the following way Propagation consistency score:

[0063] ,

[0064] in, Indicates the associated channel group The communication consistency score, Associate channel group The sum of the absolute values ​​of the propagation path integrals of all channels in , Associate channel group The Euclidean norm of the propagation path integral, Uncovering groups of associated channels The concentration of transmission, Indicates the associated channel group The variance of the propagation consistency score within Indicates the associated channel group The mean of the propagation consistency scores within Uncovering groups of associated channels The propagation dispersion.

[0065] Beneficial effects: The SEEG-based pathological high-frequency oscillation identification system provided by this solution obtains the stereoscopic EEG data of the target object (including M channel signals), pre-processes the M channel signals and then performs primary detection to determine suspicious nodes that meet the conditions; cross-channel time-frequency coherence analysis and cross-channel propagation consistency analysis are performed on each suspicious node to determine the cross-channel propagation consistency score; based on the cross-channel propagation consistency score, it is determined whether pathological high-frequency oscillations exist in the stereoscopic EEG data. This solution takes into account the characteristics of stereo EEG data (large data volume, short duration of pathological high-frequency oscillations, and irregular occurrence times), and designs a progressive architecture consisting of primary detection, cross-channel time-frequency coherence analysis, and cross-channel propagation consistency analysis. Primary detection is highly efficient and can quickly identify suspected nodes that may harbor pathological high-frequency oscillations. Preprocessing (including bandpass filtering and power frequency notching, retaining the 80-500Hz band and removing 50Hz power frequency interference) minimizes false detections caused by equipment noise (typically above 500Hz). This effectively reduces the amount of data required for in-depth processing (cross-channel time-frequency coherence analysis and cross-channel propagation consistency analysis), significantly improving operational efficiency. Leveraging the energy surge preceding pathological high-frequency oscillations (as well as physiological high-frequency oscillations), dynamic sliding window analysis is employed to calculate the channel energy within the sliding window. A threshold determination is then made to identify the target sliding window whose channel energy exceeds the dynamic energy threshold, thereby identifying suspicious nodes. This approach effectively detects nodes where pathological high-frequency oscillations may occur and avoids missed detections. Cross-channel time-frequency coherence analysis and cross-channel propagation consistency analysis enable in-depth analysis, effectively distinguishing other similar abnormal discharge signals (such as physiological high-frequency oscillations and electromyographic artifacts), and achieving high-precision detection of pathological high-frequency oscillations. Therefore, this solution provides a reliable means for high-precision detection of pathological high-frequency oscillations.

[0066] Starting from the suspicious node, a second-sized expanded window is determined to cover each channel. Time-frequency transformation is performed on the signal of each channel in the expanded window to determine the time-frequency spectrum corresponding to each channel. Based on the time-frequency spectrum corresponding to each channel, the dynamic coherence between each two channels is calculated, and adaptive processing is performed to filter out channels with low dynamic coherence and retain channels with high dynamic coherence. The dynamic coherence between each two channels is integrated to determine a three-dimensional tensor as the coherence feature corresponding to the expanded window. Because electromyographic artifacts do not have cross-channel neural propagation characteristics, by calculating the dynamic coherence between each two channels and performing adaptive processing (setting to zero if the conditions are not met), abnormal discharges caused by electromyographic artifacts can be effectively eliminated, reducing false detections caused by electromyographic artifacts. The lower coherence of physiological high-frequency oscillations than pathological high-frequency oscillations can also reduce false detections caused by physiological high-frequency oscillations to a certain extent, thereby improving the detection accuracy of pathological high-frequency oscillations.

[0067] In order to accurately identify pathological high-frequency oscillations and avoid false detections caused by physiological high-frequency oscillations, the core differences between physiological high-frequency oscillations and pathological high-frequency oscillations are considered: the cross-channel propagation ability of physiological high-frequency oscillations is not sustained (lasting less than 300ms) compared to the cross-channel propagation ability of pathological high-frequency oscillations, and is self-limited (low cross-channel coherence), while pathological high-frequency oscillations can repeatedly last for more than 500ms and can propagate across brain lobes (high cross-channel coherence). In addition, although the spatial waveform of pathological high-frequency oscillations does not have an explicit regularity during cross-channel propagation, it has explicit characteristics in propagation delay and phase (cross-channel propagation delay is 12±3ms, and has phase synchronization stability). Based on this, this scheme creatively designs cross-channel propagation consistency analysis for the accurate identification of pathological high-frequency oscillations. By The coherence eigenvalue of When the threshold is exceeded ,mark Active nodes have more than the set ratio make and At the same time, it is an active node, and the channel and channel Classify into the same associated channel group, and finally generate an associated channel group , , is the total number of associated channel groups, and for each associated channel group, the propagation path integral of each channel in the associated channel group is calculated (specifically, the channel in the associated channel group is calculated as With channel The phase difference is calculated to calculate the associated channel group Channels in With channel The propagation parameters can reveal the propagation intensity and direction, and introduce cross-channel dynamic coherence to enhance the propagation intensity and expand the difference between physiological high-frequency oscillations and pathological high-frequency oscillations), and then calculate the associated channel group Channels in The propagation path integral is used to distinguish the total outflow intensity and total inflow intensity of the channel using the half-wave rectification function, and to quantify the associated channel group. By comprehensively calculating the cross-channel propagation consistency score and judging whether the conditions are met (such as threshold conditions, the number of associated channel groups that meet the threshold conditions, etc.), accurate identification of pathological high-frequency oscillations can be achieved, effectively avoiding false detections caused by other similar abnormal discharge patterns, and effectively determining the cross-channel propagation direction, locating the propagation source (epileptic lesion) and the propagation sink (propagation path of pathological high-frequency oscillations).

[0068] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0070] Figure 1 This is an architectural diagram of the SEEG-based pathological high-frequency oscillation identification system provided in an embodiment of the present application.

[0071] Figure 2 Schematic diagram of SEEG.

[0072] Figure 3 Schematic diagram of using sliding window to perform preliminary detection on the signals of each SEEG channel.

[0073] Figure 4 Schematic diagram for identifying suspicious nodes.

[0074] Figure 5 Schematic diagram of using extended windows for cross-channel time-frequency coherence analysis and cross-channel propagation consistency analysis.

[0075] Figure 6 Schematic diagram of pathological high-frequency oscillations in SEEG.

[0076] Figure 7 Schematic diagram of pathological high-frequency oscillations in single-channel signals.

[0077] Icons: 10- SEEG-based pathological high-frequency oscillation recognition system; 11- signal acquisition module; 12- node initial inspection module; 13- scoring module; 14- decision module. DETAILED DESCRIPTION

[0078] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0079] See also Figure 1 , Figure 1 This is an architecture diagram of a SEEG-based pathological high-frequency oscillation identification system 10 provided in an embodiment of the present application. In this embodiment, the SEEG-based pathological high-frequency oscillation identification system 10 is deployed in a computer and can be functionally divided into a signal acquisition module 11, a node initial inspection module 12, a scoring module 13, and a decision module 14.

[0080] First, the signal acquisition module 11 may acquire stereoscopic EEG data of the target object, wherein the stereoscopic EEG data includes M channel signals.

[0081] In this embodiment, electrodes can be implanted into a target subject (e.g., a patient with drug-resistant focal epilepsy) and a collection device can be used to collect stereo EEG data of the target subject. The sampling frequency can be 2000 Hz and contain M channel signals. The collected stereo EEG data is as follows: Figure 2 shown.

[0082] After obtaining the stereo EEG data of the target object (partial data may be collected, and the system can perform pathological high-frequency oscillation detection while collecting data), the node initial detection module 12 pre-processes the M channel signals and performs primary detection to determine suspicious nodes that meet the conditions.

[0083] In this embodiment, the node initial detection module 12 may perform primary detection after pre-processing the M channel signals to determine whether there are suspicious nodes that meet the conditions. If there are suspicious nodes, the suspicious nodes are marked.

[0084] Exemplarily, the node initial inspection module 12 can perform bandpass filtering and power frequency trapping on the M channel signals, retaining the 80-500Hz frequency band and removing 50Hz power frequency interference. Pathological high-frequency oscillations are usually in the 80-500Hz range, while equipment noise is usually above 500Hz. Therefore, retaining 80-500Hz through bandpass filtering can minimize the interference caused by equipment noise (equipment noise is usually above 500Hz), remove 50Hz power frequency interference (if other countries or regions use 60Hz AC power, remove 60Hz power frequency interference), and eliminate power frequency harmonics (fundamental frequency 50Hz, as well as multiple harmonics such as 100Hz, 150Hz, 200Hz, and 250Hz) on stereo EEG signals.

[0085] For each channel signal after preprocessing: the node initial inspection module 12 can use a sliding window of a first size (a size of 50ms is used in this embodiment) to perform dynamic sliding window analysis on the channel signal, such as Figure 3 As shown (for easy viewing, some channels are selected for display).

[0086] Based on this, the node initial inspection module 12 can calculate the channel energy of each sliding window, determine the target sliding window whose channel energy exceeds the dynamic energy threshold, and determine the starting position of the target sliding window as a suspicious node.

[0087] Exemplarily, the node initial inspection module 12 may calculate the channel energy of each sliding window in the following manner:

[0088] , (1)

[0089] in, Indicates channel Middle The high-frequency energy index at the end of the sliding window, , For the The start time of the sliding window, ,unit , is the initial time, For the The end time of the sliding window, is the first size, , is the signal frequency, For channel The channel signal after preprocessing, is the short-time Fourier transform;

[0090] like , determine the channel Middle The sliding window is the target sliding window, and the channel is determined Middle The start time of the sliding window is a suspicious node, where satisfy:

[0091] , (2)

[0092] , (3)

[0093] , (4)

[0094] in, is the dynamic energy threshold, is the baseline energy value, express The median absolute deviation of .

[0095] Taking into account the characteristics of stereo EEG data (large data volume, short duration of pathological high-frequency oscillations, irregular occurrence time, etc.), the system designed a progressive architecture of primary detection + cross-channel time-frequency coherence analysis + cross-channel propagation consistency analysis. The primary detection has high operating efficiency and can quickly detect suspicious nodes where pathological high-frequency oscillations may occur. Preprocessing (including bandpass filtering and power frequency notching, retaining the 80-500Hz frequency band, and removing 50Hz power frequency interference) can eliminate false detections caused by equipment noise (equipment noise is usually higher than 500Hz) as much as possible. This can effectively reduce the amount of data that requires in-depth processing (cross-channel time-frequency coherence analysis + cross-channel propagation consistency analysis) and significantly improve operating efficiency. By utilizing the energy surge phenomenon before pathological high-frequency oscillations (and physiological high-frequency oscillations) occur (usually manifested as a sudden increase in signal amplitude to more than three times the baseline within 5-20 milliseconds), dynamic sliding window analysis is adopted to calculate the channel energy of the sliding window, make a threshold judgment, and determine the target sliding window whose channel energy exceeds the dynamic energy threshold, thereby identifying suspicious nodes. This can effectively detect nodes where pathological high-frequency oscillations may occur and avoid missed detections.

[0096] After the node initial inspection module 12 determines the suspicious nodes, the scoring module 13 can perform cross-channel time-frequency coherence analysis and cross-channel propagation consistency analysis on each suspicious node to determine a cross-channel propagation consistency score.

[0097] In this embodiment, for each suspicious node: the scoring module 13 determines an expansion window of a second size with the suspicious node as the starting position, wherein the second size (for example, 500ms, 1000ms, etc., 500ms is used as an example in this embodiment) is larger than the first size, and the expansion window covers each channel, such as Figure 5 As shown (the green box represents an expanded window). It should be noted that, to avoid missed detections, after determining a suspicious node, the scoring module 13 may define a depth detection window based on the suspicious node (e.g., starting from the suspicious node and lasting for 5 seconds, 10 seconds, 30 seconds, etc.). Within this depth detection window, a sliding window analysis is performed using an expanded window of the second size, with a step size of 100ms.

[0098] For each extended window, the scoring module 13 can perform cross-channel time-frequency coherence analysis on each channel signal in the extended window to determine the coherence feature corresponding to the extended window.

[0099] Exemplarily, the scoring module 13 may first perform a time-frequency transformation on each channel signal in the expansion window to determine a time-frequency spectrum corresponding to each channel.

[0100] For each channel signal in the extended window, the scoring module 13 can calculate the time-frequency spectrum corresponding to the channel in the following manner:

[0101] , (5)

[0102] wherein, denotes a channel the corresponding time-frequency spectrum, is a time point, , is a second dimension, is a first dimension, is a start time of the m-th sliding window, i.e. a start time of the extended window, is a signal frequency, , is a channel the pre-processed channel signal, is an integration variable, is a Morlet wavelet, is a scale parameter, denotes a complex conjugate.

[0103] According to the determination of the corresponding time-frequency spectrum of each channel in the extended window, the scoring module 13 can calculate the dynamic coherence between each two channels based on the corresponding time-frequency spectrum of each channel.

[0104] For each two channels, the scoring module 13 can calculate the dynamic coherence between the two channels in the following manner:

[0105] , (6)

[0106] wherein, denotes a channel the dynamic coherence between channel at time point and frequency , , is a time index, , , is a wavelet coefficient, is a complex conjugate of the wavelet coefficient.

[0107] According to the determination of the dynamic coherence between each two channels in the extended window, the scoring module 13 can further process the dynamic coherence between each two channels to determine the coherence feature corresponding to the extended window.

[0108] For the dynamic coherence between each two channels, the scoring module 13 can perform adaptive processing in the following manner:

[0109] , (7)

[0110] wherein, the channels after adaptive processing between the channels at a time point and a frequency , the average of the dynamic coherence between the channels in the extended window between the channels , the standard deviation of the dynamic coherence between the channels ,

[0111] After that, the scoring module 13 can integrate the dynamic coherence between each two channels after adaptive processing at each time point in the extended window to obtain the coherence feature corresponding to the extended window:

[0112] , (8)

[0113] wherein, represents the coherence feature value of the channel at the time point and the frequency , represents the maximum value of the dynamic coherence between the channel and all channels except the channel , represents the operation on each channel to generate a three-dimensional data structure containing all channels, is the length of the frequency dimension, represents a three-dimensional tensor.

[0114] The second size of the extended window is determined starting from the suspicious node to cover each channel. Time-frequency transformation is performed on each channel signal in the extended window to determine the corresponding time-frequency spectrum of each channel. Based on the corresponding time-frequency spectrum of each channel, the dynamic coherence between each two channels is calculated, and adaptive processing is performed to exclude channels with low dynamic coherence and retain channels with high dynamic coherence. The dynamic coherence between each two channels is integrated to determine a three-dimensional tensor as the coherence feature corresponding to the extended window. Since the electromyographic artifact does not have the neural transmission characteristic across channels, by calculating the dynamic coherence between each two channels and performing adaptive processing (those that do not meet the conditions are set to zero), the abnormal discharge phenomenon caused by the electromyographic artifact can be effectively excluded, reducing the false detection caused by the electromyographic artifact. The coherence of physiological high-frequency oscillation is lower than that of pathological high-frequency oscillation, which can also reduce the false detection caused by physiological high-frequency oscillation to a certain extent, thereby improving the detection accuracy of pathological high-frequency oscillation.

[0115] To accurately identify pathological high-frequency oscillations and avoid false detections caused by physiological high-frequency oscillations, the core differences between physiological and pathological high-frequency oscillations are considered: the cross-channel propagation ability of physiological high-frequency oscillations is not sustained (lasting less than 300ms) and is self-limited (low cross-channel coherence), compared to the cross-channel propagation ability of pathological high-frequency oscillations. Pathological high-frequency oscillations, on the other hand, can repeatedly persist for more than 500ms and propagate across brain lobes (high cross-channel coherence). Furthermore, although the spatial waveform of pathological high-frequency oscillations lacks explicit regularity during cross-channel propagation, they do have explicit characteristics in propagation delay and phase (cross-channel propagation delay is 12±3ms, and phase synchronization stability is demonstrated). Based on this, the system creatively designed cross-channel propagation consistency analysis to accurately identify pathological high-frequency oscillations.

[0116] After obtaining the coherence feature corresponding to the extended window, the scoring module 13 may perform cross-channel propagation consistency analysis on the channel signals in the extended window based on the coherence feature corresponding to the extended window to determine a cross-channel propagation consistency score.

[0117] In this embodiment, for each time point in the extended window and frequency , if the channel The coherence eigenvalue of Exceeding the threshold , scoring module 13 marks is an active node. And for any channel and channel If there is more than the set ratio make and At the same time, the scoring module 13 will be an active node. and channel Classify them into the same associated channel group, and finally generate an associated channel group:

[0118] , (9)

[0119] in, , is the total number of associated channel groups.

[0120] For each associated channel group, the scoring module 13 may calculate the propagation path integral of each channel in the associated channel group, and determine a cross-channel propagation consistency score based on the propagation path integral.

[0121] For example, the scoring module 13 may calculate the channels in the associated channel group in the following manner: With channel Phase difference:

[0122] , (10)

[0123] , (11)

[0124] in, Associate channel group Channels in With channel At the time point The instantaneous phase difference, For time point, , For the second size, For the The start time of the sliding window, that is, the start time of the extended window, For channel The analytical signal, For channel The complex conjugate of the analytical signal, represents the argument of a complex number, For channel The channel signal after preprocessing, is an imaginary number, is the Hilbert transform;

[0125] Calculate the associated channel group After calculating the phase difference between each two channels, the scoring module 13 can calculate the associated channel group in the following way Channels in With channel Propagation parameters:

[0126] , (12)

[0127] in, Indicates the associated channel group Channels in Towards channel The propagation intensity of directional propagation, Instantaneous phase difference The time gradient of is a symbolic function that reveals the channel With channel The direction of propagation between

[0128] Calculate the associated channel group After the propagation parameters of each two channels are obtained, the scoring module 13 can calculate the associated channel group in the following way Channels in The propagation path integral of :

[0129] , (13)

[0130] in, Indicates the associated channel group Channels in The propagation path integral reveals the channel The net transmission intensity, Associate channel group The number of channels in is a half-wave rectifier function, retaining only positive values, Indicates the associated channel group Channels in The total outflow intensity, Indicates the associated channel group Channels in Total inflow intensity;

[0131] Finally, the scoring module 13 can calculate the associated channel group in the following way Propagation consistency score:

[0132] , (14)

[0133] in, Indicates the associated channel group The communication consistency score, Associate channel group The sum of the absolute values ​​of the propagation path integrals of all channels in , Associate channel group The Euclidean norm of the propagation path integral, Uncovering groups of associated channels The concentration of transmission, Indicates the associated channel group The variance of the propagation consistency score within Indicates the associated channel group The mean of the propagation consistency scores within Uncovering groups of associated channels The propagation dispersion.

[0134] Through the channel The coherence eigenvalue of Exceeding the threshold When, mark Active nodes have more than the set ratio make and At the same time, it is an active node, and the channel and channel Classify into the same associated channel group, and finally generate an associated channel group , , is the total number of associated channel groups, and for each associated channel group, the propagation path integral of each channel in the associated channel group is calculated (specifically, the channel in the associated channel group is calculated as With channel The phase difference is calculated to calculate the associated channel group Channels in With channel The propagation parameters can reveal the propagation intensity and direction, and introduce cross-channel dynamic coherence to enhance the propagation intensity and expand the difference between physiological high-frequency oscillations and pathological high-frequency oscillations), and then calculate the associated channel group Channels in The propagation path integral is used to distinguish the total outflow intensity and total inflow intensity of the channel using the half-wave rectification function, and to quantify the associated channel group. By comprehensively calculating the cross-channel propagation consistency score and judging whether the conditions are met (such as threshold conditions, the number of associated channel groups that meet the threshold conditions, etc.), accurate identification of pathological high-frequency oscillations can be achieved, effectively avoiding false detections caused by other similar abnormal discharge patterns, and effectively determining the cross-channel propagation direction, locating the propagation source (epileptic lesion) and the propagation sink (propagation path of pathological high-frequency oscillations).

[0135] Based on this, the scoring module 13 can calculate the propagation consistency score of each associated channel group. Then, the decision module 14 can determine whether pathological high-frequency oscillations exist in the stereo EEG data based on the cross-channel propagation consistency score.

[0136] In this embodiment, the decision module 14 may use the propagation consistency score of the associated channel group to make a judgment:

[0137] If the highest score of the propagation consistency scores of all associated channel groups exceeds the score threshold (e.g., 1.25), the decision module 14 determines that pathological high-frequency oscillations exist in the current expansion window, e.g. Figure 6 and Figure 7 As shown, Figure 6 The blue vertical line marks the starting point of pathological high-frequency oscillation. Figure 7Demonstrating pathological high-frequency oscillations in a single channel. If the highest propagation consistency score of all associated channel groups does not exceed the scoring threshold, but a set number (e.g., three) of these scores exceed another scoring threshold (e.g., 1.10), decision module 14 also determines that pathological high-frequency oscillations exist in the current extended window. Otherwise, decision module 14 deems that pathological high-frequency oscillations do not exist in the current extended window, and the system proceeds to perform cross-channel time-frequency coherence analysis and cross-channel propagation consistency analysis for the next extended window (within the depth detection window). If the depth detection window has been completed and pathological high-frequency oscillations do not exist in the last extended window, node preliminary inspection module 12 continues preliminary inspection starting from the end of the depth detection window. If the depth detection window has been completed but pathological high-frequency oscillations still exist in the last extended window, scoring module 13 generates the next depth detection window starting from the end of the depth detection window and continues cross-channel time-frequency coherence analysis and cross-channel propagation consistency analysis for the next extended window in the next depth detection window.

[0138] Generally speaking, the deep detection window is sufficient to cover the duration of the onset of the pathological high-frequency oscillation. Therefore, after the deep detection window is completed, the pathological high-frequency oscillations detected by the deep detection window can be merged to generate a pathological high-frequency oscillation mark, and can be located (can be specific to the extended window level) and traced. The propagation source and propagation sink can be determined by the propagation consistency score. The highest propagation consistency score is usually the propagation source, and the rest are propagation sinks, but there may also be a small number of unexpected situations. In addition, combined with the propagation parameters between channels, the propagation path can be analyzed and determined.

[0139] In summary, the embodiment of the present application provides a pathological high-frequency oscillation identification system 10 based on SEEG, which obtains stereoscopic EEG data of the target object (including M channel signals), pre-processes the M channel signals and then performs primary detection to determine suspicious nodes that meet the conditions; performs cross-channel time-frequency coherence analysis and cross-channel propagation consistency analysis on each suspicious node to determine a cross-channel propagation consistency score; based on the cross-channel propagation consistency score, determines whether pathological high-frequency oscillations exist in the stereoscopic EEG data. This solution takes into account the characteristics of stereo EEG data (large data volume, short duration of pathological high-frequency oscillations, and irregular occurrence times), and designs a progressive architecture consisting of primary detection, cross-channel time-frequency coherence analysis, and cross-channel propagation consistency analysis. Primary detection is highly efficient and can quickly identify suspected nodes that may harbor pathological high-frequency oscillations. Preprocessing (including bandpass filtering and power frequency notching, retaining the 80-500Hz band and removing 50Hz power frequency interference) minimizes false detections caused by equipment noise (typically above 500Hz). This effectively reduces the amount of data required for in-depth processing (cross-channel time-frequency coherence analysis and cross-channel propagation consistency analysis), significantly improving operational efficiency. Leveraging the energy surge preceding pathological high-frequency oscillations (as well as physiological high-frequency oscillations), dynamic sliding window analysis is employed to calculate the channel energy within the sliding window. A threshold determination is then made to identify the target sliding window whose channel energy exceeds the dynamic energy threshold, thereby identifying suspicious nodes. This approach effectively detects nodes where pathological high-frequency oscillations may occur and avoids missed detections. Cross-channel time-frequency coherence analysis and cross-channel propagation consistency analysis can perform in-depth analysis and effectively distinguish other similar abnormal discharge signals (such as physiological high-frequency oscillations and electromyographic artifacts), thereby achieving high-precision detection of pathological high-frequency oscillations.

[0140] In this document, relational terms such as first and second, etc. are used merely to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any actual relationship or order between these entities or operations.

[0141] The above description is merely an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, various modifications and variations of the present application are possible. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A pathological high-frequency oscillation recognition system based on SEEG, characterized in that: include: A signal acquisition module is used to acquire stereoscopic EEG data of a target object, wherein the stereoscopic EEG data includes M channel signals; The node initial inspection module is used to perform primary inspection after pre-processing the M channel signals to determine the suspicious nodes that meet the conditions; The scoring module is used to perform cross-channel time-frequency coherence analysis and cross-channel propagation consistency analysis on each suspicious node to determine the cross-channel propagation consistency score; a decision module for determining whether pathological high-frequency oscillations exist in stereo EEG data based on a cross-channel propagation consistency score; The scoring module is specifically used for: For each suspicious node: determining an expansion window of a second size with the suspicious node as a starting position, wherein the second size is larger than the first size, and the expansion window covers each channel; For each extended window: perform cross-channel time-frequency coherence analysis on each channel signal in the extended window to determine the coherence characteristics corresponding to the extended window; based on the coherence characteristics corresponding to the extended window, perform cross-channel propagation consistency analysis on the channel signals in the extended window to determine the cross-channel propagation consistency score.

2. The pathological high-frequency oscillation identification system based on SEEG according to claim 1, characterized in that: The node initial inspection module is specifically used for: Perform bandpass filtering and power frequency notching on the M channel signals, retaining the 80-500Hz frequency band and removing the 50Hz power frequency interference; For each channel signal after preprocessing: A sliding window of the first size is used to perform dynamic sliding window analysis on the channel signal, the channel energy of each sliding window is calculated, the target sliding window whose channel energy exceeds the dynamic energy threshold is determined, and the starting position of the target sliding window is determined as a suspicious node.

3. The pathological high-frequency oscillation identification system based on SEEG according to claim 2, characterized in that: The node initial inspection module is specifically used for: The channel energy of each sliding window is calculated as follows: ， in, represents the high-frequency energy index at the end of the k-th sliding window in channel c, c∈[1,M], is the start time of the kth sliding window, , unit ms, is the initial time, is the end time of the kth sliding window, is the first size, , is the signal frequency, is the channel signal after channel c preprocessing, STFT is short-time Fourier transform; like , determine the channel Middle The sliding window is the target sliding window, and the channel is determined Middle The start time of the sliding window is a suspicious node, where satisfy: ， ， ， in, is the dynamic energy threshold, is the baseline energy value, express The median absolute deviation of .

4. The pathological high-frequency oscillation identification system based on SEEG according to claim 1, characterized in that: The scoring module is specifically used to: Perform time-frequency transformation on each channel signal in the expansion window to determine the time-frequency spectrum corresponding to each channel; Based on the time-frequency spectrum corresponding to each channel, the dynamic coherence between every two channels is calculated; The dynamic coherence between every two channels is processed to determine the coherence characteristics corresponding to the expanded window.

5. The pathological high-frequency oscillation identification system based on SEEG according to claim 4, characterized in that: The scoring module is specifically used to: For each channel signal in the expansion window, the time-frequency spectrum corresponding to the channel is calculated in the following way: ， in, Indicates channel The corresponding time-frequency spectrum is For time point, , For the second size, For the The start time of the sliding window, that is, the start time of the extended window, is the signal frequency, , For channel The channel signal after preprocessing, is the integration variable, is the Morlet wavelet, is the scale parameter, represents the complex conjugate.

6. The pathological high-frequency oscillation identification system based on SEEG according to claim 5, characterized in that: The scoring module is specifically used to: For every two channels, the dynamic coherence between the two channels is calculated as follows: ， in, Indicates channel With channel Between the time points and frequency Dynamic coherence on , is the time index, , 、 is the wavelet coefficient, is the complex conjugate of the wavelet coefficient.

7. The pathological high-frequency oscillation identification system based on SEEG according to claim 6, characterized in that: The scoring module is specifically used to: For the dynamic coherence between every two channels, adaptive processing is performed in the following way: ， in, The channel after adaptive processing With channel Between the time points and frequency The dynamic coherence on Channels in the extended window With channel The mean dynamic coherence between For channel With channel The standard deviation of the dynamic coherence between The dynamic coherence between each two channels after adaptive processing at each time point in the integrated expansion window , and get the coherence characteristics corresponding to the expanded window: ， in, Indicates channel At the time point and frequency The coherence eigenvalue when Indicates that the channel All channels except Take and Channel The maximum value of the dynamic coherence, For each channel Perform operations to generate a three-dimensional data structure containing all channels. is the length of the frequency dimension, Represents a three-dimensional tensor.

8. The pathological high-frequency oscillation identification system based on SEEG according to claim 7, characterized in that: The scoring module is specifically used to: For each time point in the extended window and frequency , if the channel The coherence eigenvalue of Exceeding the threshold ,mark is an active node; For any channel and channel If there is more than the set ratio make and At the same time, it is an active node, and the channel and channel Classify into the same associated channel group, and finally generate an associated channel group , , is the total number of associated channel groups; For each associated channel group: calculate the propagation path integral of each channel in the associated channel group, and determine the cross-channel propagation consistency score based on the propagation path integral.

9. The pathological high-frequency oscillation identification system based on SEEG according to claim 8, characterized in that: The scoring module is specifically used to: The channels in the associated channel group are calculated in the following way With channel Phase difference: ， ， in, Associate channel group Channels in With channel At the time point The instantaneous phase difference, For time point, , For the second size, For the The start time of the sliding window, that is, the start time of the extended window, For channel The analytical signal, For channel The complex conjugate of the analytical signal, represents the argument of a complex number, For channel The channel signal after preprocessing, is an imaginary number, is the Hilbert transform; The associated channel group is calculated in the following way Channels in With channel Propagation parameters: ， in, Indicates the associated channel group Channels in Towards channel The propagation intensity of directional propagation, Instantaneous phase difference The time gradient of is a symbolic function that reveals the channel With channel The direction of propagation between The associated channel group is calculated in the following way Channels in The propagation path integral of : ， in, Indicates the associated channel group Channels in The propagation path integral reveals the channel The net transmission intensity, Associate channel group The number of channels in is a half-wave rectifier function, retaining only positive values, Indicates the associated channel group Channels in The total outflow intensity, Indicates the associated channel group Channels in Total inflow intensity; The associated channel group is calculated in the following way Propagation consistency score: ， in, Indicates the associated channel group The communication consistency score, Associate channel group The sum of the absolute values ​​of the propagation path integrals of all channels in , Associate channel group The Euclidean norm of the propagation path integral, Uncovering groups of associated channels The concentration of transmission, Indicates the associated channel group The variance of the propagation consistency score within Indicates the associated channel group The mean of the propagation consistency scores within Uncovering groups of associated channels The propagation dispersion.

Images