Ripple wave signal detection and identification system based on IED events and time-frequency features

Abstract

The application provides a ripples wave signal detection and recognition system based on IED events and time-frequency characteristics, comprising: a signal acquisition module acquires electroencephalogram data of a target object, a preprocessing module pre-processes each channel signal of the electroencephalogram data, including power frequency notch filtering and band-pass filtering, a 25-80 Hz frequency band is a first frequency band signal, and a 70-180 Hz frequency band is a second frequency band signal; an IED detection module performs IED signal detection on the first frequency band signal to determine an IED event node; a ripples wave detection module performs sliding window analysis on the second frequency band signal to perform ripples wave signal detection and determine a ripples wave event node; and a decision recognition module determines a ripples wave signal category based on the IED event node and the ripples wave event node. By detecting IED events and ripples wave events, the IED events are introduced into the ripples wave event classification, and the physiological ripples wave and pathological signals are automatically detected and recognized in combination with frequency domain characteristics.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of electroencephalogram signal research, in particular to a ripple wave signal detection and identification system based on IED events and time-frequency characteristics. BACKGROUND

[0002] In the field of electroencephalogram neuroscience, ripple and IED (Interictal Epileptiform Discharges) are core pathological signals. Ripple is defined as a specific 80-200 Hz high-frequency oscillation, mainly originating from the hippocampus and the temporal cortex. Physiological ripple is defined as a signal that appears in slow wave sleep and drives information encoding and storage in the memory consolidation process; pathological signal is highly associated with seizures, and its abnormal enhancement can activate the epileptogenic network, which is a key biomarker for locating the epileptogenic focus. Ripple detection requires intracranial electrodes (such as deep electrodes or ECoG) and a sampling rate of more than 2000 Hz.

[0003] IED refers to characteristic transient waveforms of brain electrical activity that occur between seizures in patients with epilepsy. Typical forms include spike waves (20-70 ms high amplitude sharp waves), sharp waves, and spike slow complex waves (spike waves followed by 200-500 ms slow waves). These discharges do not directly trigger overt seizures, but reflect abnormal cortical hypersynchrony and are the gold standard for diagnosing epilepsy - its spatial distribution can indicate the location of the epileptogenic zone for resection surgery, and its frequency change can evaluate the efficacy of antiepileptic drugs. In clinical practice, scalp EEG has a sensitivity of about 30% for superficial IED in the cortex, and stereotactic electroencephalography (SEEG) technology is required to accurately capture deep lesions.

[0004] Pathological signals often coexist with IED events: high-frequency oscillations can be attached to the spike component of IED, and this coupling phenomenon indicates the exacerbation of neuronal cluster hypersynchrony, with a sensitivity of up to 90% for predicting seizures.

[0005] The previous research of the research team (Hippocampal ripples correlate with memory performance in humans, Qing-Tian Duan, Lu Dai, Lu-Kang Wang, Xian-Jun Shi, Xiaowei Chen, Xiang Liao, Chun-Qing Zhang, Hui Yang) shows that in the hippocampal region (CA1-CA4, Subiculum, Presubiculum, DG, CA1-CA4 represents hippocampal angle 1-4 area, Subiculum represents lower support, Presubiculum represents front lower support, and DG represents dentate gyrus), the ripple mainly concentrates in 70-180Hz, the physiological ripple has a positive effect on memory, and the pathological signal and the IED event have a negative effect on memory, and both have a significant correlation.

[0006] Therefore, developing a signal detection system capable of automatically identifying physiological ripples and pathological signals can provide a strong aid for studying and evaluating memory disorders, and therefore, the team develops a ripple signal detection and identification system based on IED events and time-frequency characteristics. SUMMARY

[0007] The purpose of the embodiment of the present application is to provide a ripple signal detection and identification system based on IED events and time-frequency characteristics, by detecting IED events and ripple events, introducing IED events in the classification of ripple events, and combining frequency domain characteristics to realize automatic detection and identification of physiological ripples and pathological signals.

[0008] In order to achieve the above-mentioned purpose, the embodiments of the present application are realized by the following ways:

[0009] In a first aspect, the embodiments of the present application provide a ripple wave signal detection and identification system based on IED events and time-frequency characteristics, comprising: a signal acquisition module configured to acquire electroencephalogram data of a target object, wherein the electroencephalogram data at least contains channel signals of each hippocampal subregion, the hippocampal subregion includes CA1-CA4, Subiculum, Presubiculum and DG, CA1-CA4 represents hippocampal cornu ammonis 1-4 region, Subiculum represents subiculum, Presubiculum represents presubiculum, and DG represents dentate gyrus; a preprocessing module configured to preprocess each channel signal of the electroencephalogram data, wherein the preprocessing includes power frequency notch filtering and band-pass filtering, the power frequency notch filtering removes 50Hz power frequency interference, the band-pass filtering retains 25-80Hz frequency band and 70-180Hz frequency band respectively, the 25-80Hz frequency band is a first frequency band signal, and the 70-180Hz frequency band is a second frequency band signal; an IED detection module configured to perform IED signal detection on the first frequency band signal to determine an IED event node, wherein IED represents interictal epileptiform discharge; a ripple wave detection module configured to perform sliding window analysis on the second frequency band signal to perform ripple wave signal detection and determine a ripple wave event node; and a decision identification module configured to determine a ripple wave signal category based on the IED event node and the ripple wave event node, wherein the ripple wave signal category includes physiological ripple wave and pathological signal.

[0010] In combination with the first aspect, in a first possible implementation manner of the first aspect, the IED detection module is specifically configured to: for the first frequency band signal of each channel: perform global analysis on the first frequency band signal of the current channel to calculate a baseline voltage of the current channel; take the baseline voltage as a reference to perform peak value detection on the first frequency band signal of the current channel and determine a candidate node meeting a peak value condition, wherein the peak value condition is that higher than a sum of the baseline voltage and 3 times of MAD, and MAD represents median absolute deviation; for each candidate node, take the candidate node as a reference to determine a 200ms interval forward and a 400ms interval backward to obtain a 600ms time window corresponding to the candidate node, then perform spike waveform verification, slow wave oscillation verification and background suppression verification on signals in the time window, and determine a marked candidate node meeting requirements based on verification results of the spike waveform verification, the slow wave oscillation verification and the background suppression verification, and determine an IED event starting point and an IED event ending point based on the marked candidate node.

[0011] In conjunction with the first possible implementation of the first aspect, in the second possible implementation of the first aspect, the IED detection module is specifically used for: determining the first valley point after the peak point within the time window based on the peak point corresponding to the candidate node, determining the first 40% valley amplitude point before and after the valley point, calculating the spike morphology symmetry index, and determining the spike morphology verification result; determining the first valley point after the peak point within the time window based on the peak point corresponding to the candidate node, then determining the two valley points after the first valley point, and determining the oscillation attenuation verification result based on the value of each valley point; determining the peak point before the peak point within the time window based on the peak point corresponding to the candidate node, and using the previous peak point as the endpoint, determining the background suppression verification region within the time window, and determining the background suppression verification result.

[0012] In conjunction with the first aspect, in the third possible implementation of the first aspect, the ripple wave detection module is specifically used for: for the second frequency band signal of each channel: performing global analysis on the second frequency band signal of the current channel to calculate the baseline voltage of the current channel; performing sliding window processing on the second frequency band signal of the current channel to obtain several window data, wherein the sliding window size is 100ms and the step size is 20ms; for each window data: performing feature extraction on the window data to determine several feature indices, and judging based on the feature indices to determine candidate ripple wave events, and then determining the start node and end node of each candidate ripple wave event to obtain several ripple wave event nodes.

[0013] In conjunction with the third possible implementation of the first aspect, in the fourth possible implementation of the first aspect, the ripple wave detection module is specifically used for: calculating the energy envelope of the window data to obtain the window energy envelope; counting the number of peaks in the window data; determining whether there are three or more consecutive window data whose window energy envelopes are greater than the sum of the baseline voltage of the front channel and four times the MAD, and the number of peaks in each window data is not less than eight. If so, the consecutive window data are merged as a candidate ripple wave event, where MAD represents the median absolute deviation.

[0014] In conjunction with the fourth possible implementation of the first aspect, in the fifth possible implementation of the first aspect, the ripple wave detection module is specifically used to calculate the window energy envelope in the following manner:

[0015] ,

[0016] in, The window energy envelope of the window data. The number of sampling points for the window data. For the first data in the window The sampled voltage values ​​of each data point.

[0017] In conjunction with the fourth possible implementation of the first aspect, in the sixth possible implementation of the first aspect, the ripple wave detection module is specifically used for: for each candidate ripple wave event: determining the time node from the first window data of the candidate ripple wave event when the voltage amplitude first exceeds the sum of the baseline voltage and twice the MAD, as the starting node of the candidate ripple wave event; and determining the time node from the last window data of the candidate ripple wave event when the voltage amplitude last exceeds the sum of the baseline voltage and twice the MAD, as the ending node of the candidate ripple wave event.

[0018] In conjunction with the second possible implementation of the first aspect, in the seventh possible implementation of the first aspect, the decision identification module is specifically used for: For each channel: performing event node matching based on IED event nodes and ripple wave event nodes, determining that ripple wave event nodes whose starting node is located within a target time window are pathological signals, where the target time window represents a 50ms time window starting from the peak point corresponding to the IED event node; For ripple wave event nodes in each channel that are not determined to be pathological signals: performing rapid spectrum analysis and energy concentration analysis on the merged window data where the ripple wave event nodes are located, determining the spectral peak frequency and energy concentration, and then determining whether the ripple wave event nodes are pathological signals based on the spectral peak frequency and energy concentration.

[0019] In conjunction with the seventh possible implementation of the first aspect, in the eighth possible implementation of the first aspect, the decision identification module is specifically used for: performing rapid spectral analysis on the merged window data where the ripple wave event nodes are located, and calculating the spectral peak frequencies.

[0020] ,

[0021] in, To merge the spectral peak frequencies of the window data, For frequency, Fourier transform of the merged window data; energy concentration analysis of the merged window data containing the ripple wave event nodes, and calculation of energy concentration:

[0022] ,

[0023] in, To optimize the energy concentration of merged window data, for ~ Energy integral of frequency band, The energy integral is the energy across the entire frequency band from 70 to 180 Hz.

[0024] In a ninth possible implementation of the first aspect, in combination with the seventh possible implementation of the first aspect, the decision recognition module is specifically configured to: if the spectral peak frequency of the merged window data where the ripples event node is located is not less than 140 Hz, and the energy concentration degree is not less than 0.35, determine that the ripples event node is a pathological signal; if the spectral peak frequency of the merged window data where the ripples event node is located is not more than 120 Hz, and the energy concentration degree is not more than 0.2, determine that the ripples event node is a physiological ripple; if the spectral peak frequency of the merged window data where the ripples event node is located is within (120, 140), and the energy concentration degree is within (0.2, 0.35), calculate the probability that the ripples event node belongs to a pathological signal based on the spectral peak frequency and the energy concentration degree, and determine whether the ripples event node is a pathological signal.

[0025] Beneficial effects:

[0026] The system for detecting and identifying ripples signals based on IED events and time-frequency characteristics provided in the scheme acquires the EEG data of the target object (including the channel signals of each hippocampal subregion, and the hippocampal subregions include CA1-CA4, Subiculum, Presubiculum, and DG) through the signal acquisition module. The power frequency notch and band-pass filtering are performed on each channel signal of the EEG data by the preprocessing module, the power frequency notch removes the 50 Hz power frequency interference, and the band-pass filtering retains the first frequency band signal of the 25-80 Hz frequency band and the second frequency band signal of the 70-180 Hz frequency band, respectively. The IED signal detection is performed on the first frequency band signal by the IED detection module to determine the IED event node, the sliding window analysis is performed on the second frequency band signal by the ripples detection module to detect the ripples signal, and the ripples event node is determined. Then, the decision recognition module determines the ripples signal category (physiological ripples or pathological signal) based on the IED event node and the ripples event node. The system constructs the detection basis of time-space coupling through the hippocampal subregion partition acquisition of the signal acquisition module and the dual-frequency band separation (25-80 Hz and 70-180 Hz) of the preprocessing module. The IED event is taken as the time-space anchor point of the pathological signal (the specific construction target time window is taken as the judgment basis) to assist the recognition and classification of the pathological signal. The essential difference between the physiological ripples and the pathological signal is analyzed for the ripples event not coupled with the IED, and the dual-index discrimination of the spectral peak frequency and the energy concentration degree is designed to essentially distinguish the pathological signal and the physiological ripples. The pathological signal appearing independently can be effectively recognized, the accuracy of the ripples category division is improved, and the misclassification probability is reduced.

[0027] In the identification of IED events, by analyzing the waveform shape at the occurrence of IED, the detection mechanism of IED event is designed, and a variety of discriminant standards (based on the peak point corresponding to the candidate node, the first valley point is determined after the peak point in the time window, the first 40% valley value points before and after the valley point are determined, the symmetry index of the spike waveform is calculated, and the spike waveform verification result is determined; based on the peak point corresponding to the candidate node, the first valley point is determined after the peak point in the time window, and then the first two valley points after the first valley point are determined, and the oscillation decay verification result is determined based on the value of each valley point; based on the peak point corresponding to the candidate node, the peak point before the peak point in the time window is determined, and the background suppression verification area is determined with the previous peak point as the end point, and the background suppression verification result is determined) are superimposed, which can effectively deal with various typical forms of IED events (including spike wave, sharp wave, spike slow complex wave, etc.), improve the accuracy of IED event identification, and based on the coupling window characteristics of pathological signals and IED events in the previous research results, the coupling standard of IED event node as pathological signal (50ms time window with the peak point corresponding to the IED event node as the starting point) is determined, which improves the identification accuracy of pathological signals coupled with IED events.

[0028] 3. In the process of detecting ripples, the baseline voltage is calculated by performing global analysis on the second frequency band signal, and then the sliding window processing is performed (the sliding window size is 100ms, and the step is 20ms, which can consider the short-term characteristics of physiological ripples, improve the detection accuracy of ripple events, and reduce the false negative rate), and a plurality of window data are obtained. For each window data: the window energy envelope is obtained by performing energy envelope calculation on the window data; the number of wave peaks of the window data is counted; it is judged whether there are continuous 3 or more window data whose window energy envelope is greater than the sum of the baseline voltage of the previous channel and 4 times the MAD (median absolute deviation), and the number of wave peaks of each window data is not less than 8, if there are, the continuous window data are merged as a candidate ripple event. And from the first window data of the candidate ripple event, the time node at which the voltage amplitude first exceeds the sum of the baseline voltage and 2 times the MAD is determined as the starting node of the candidate ripple event; from the last window data of the candidate ripple event, the time node at which the voltage amplitude last exceeds the sum of the baseline voltage and 2 times the MAD is determined as the termination node of the candidate ripple event. This method not only can effectively detect ripple events, but also can relatively accurately determine the starting range of the ripple event, so as to improve the discrimination accuracy and reduce the false positive probability (i.e. physiological ripples are misidentified as pathological signals) in the subsequent IED event coupling discrimination process.

[0029] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the following preferred embodiments are specifically described below, and the accompanying drawings are shown as follows. Attached Figure Description

[0030] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a schematic diagram of the framework of the ripple wave signal detection and recognition system based on IED events and time-frequency characteristics provided in the embodiments of this application.

[0032] Figure 2 This is a signal diagram for the first frequency band, 25-80Hz.

[0033] Figure 3 This is a signal diagram for the second frequency band, 70-180Hz.

[0034] Icons: 10-Ripple wave signal detection and recognition system based on IED event and time-frequency characteristics; 11-Signal acquisition module; 12-Preprocessing module; 13-IED detection module; 14-Ripple wave detection module; 15-Decision recognition module. Detailed Implementation

[0035] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0036] like Figure 1 As shown, Figure 1 A schematic diagram of the framework of the ripple wave signal detection and recognition system 10 based on IED events and time-frequency characteristics provided in the embodiments of this application.

[0037] In this embodiment, the ripple wave signal detection and recognition system 10 based on IED events and time-frequency characteristics may include a signal acquisition module 11, a preprocessing module 12, an IED detection module 13, a ripple wave detection module 14, and a decision recognition module 15.

[0038] Since the present scheme mainly studies each hippocampal subfield (the hippocampal subfield includes CA1-CA4, Subiculum, Presubiculum, DG, CA1-CA4 represents hippocampal horn 1-4 area, Subiculum represents the lower support, Presubiculum represents the front lower support, and DG represents the dentate gyrus), the sampling rate requirement is at least 2000Hz, the scalp electroencephalogram sampling frequency is generally 500-2000Hz, the intracranial electroencephalogram sampling frequency is generally 2000-5000Hz, and the stereo electroencephalogram is generally 2000-10000Hz, all of which meet the requirements. However, due to the difference in collection methods, the signal intensity of the collected electroencephalogram data for hippocampal subfield positioning is different, and the difficulty (or accuracy) of hippocampal subfield positioning is scalp electroencephalogram > intracranial electroencephalogram > stereo electroencephalogram, and the signal intensity is scalp electroencephalogram < intracranial electroencephalogram < stereo electroencephalogram. However, the scalp electroencephalogram does not need to intervene in the brain, the intracranial electroencephalogram needs to place the electrode array directly on the surface of the brain cortex, and the stereo electroencephalogram needs to vertically implant the deep electrode into the brain parenchyma through the stereotactic technique, so from the convenience of implementation conditions, the scalp electroencephalogram > intracranial electroencephalogram > stereo electroencephalogram. The present embodiment does not limit which electroencephalogram data to use, and the present embodiment takes 2000Hz stereo electroencephalogram as an example for illustration.

[0039] In the present embodiment, the signal acquisition module 11 can acquire the electroencephalogram data of the target object, and the electroencephalogram data at least contains the channel signals of each hippocampal subfield (each hippocampal subfield can include more than one electrode collected channel signal), and the hippocampal subfield includes CA1-CA4, Subiculum, Presubiculum, DG, CA1-CA4 represents hippocampal horn 1-4 area, Subiculum represents the lower support, Presubiculum represents the front lower support, and DG represents the dentate gyrus.

[0040] After obtaining the electroencephalogram data of the target object, the preprocessing module 12 can preprocess each channel signal of the electroencephalogram data. Specifically, the preprocessing can include power frequency notch and band pass filtering, the power frequency notch removes 50Hz power frequency interference, and the band pass filtering respectively retains 25-80Hz frequency band and 70-180Hz frequency band, 25-80Hz frequency band is recorded as first frequency band signal, and 70-180Hz frequency band is recorded as second frequency band signal, as shown in Figure 2 and Figure 3 , wherein, Figure 2 and Figure 3 The horizontal axis of the above formula represents time (unit s, second), and the vertical axis represents voltage (unit uv, microvolt).

[0041] After preprocessing each channel signal, the first frequency band signal and the second frequency band signal corresponding to each channel can be obtained.

[0042] Accordingly, the IED detection module 13 can perform IED signal detection on the first frequency band signal to determine an IED event node, where IED represents inter-ictal epileptiform discharge.

[0043] In this embodiment, for the first frequency band signal of each channel: the IED detection module 13 can perform global analysis on the first frequency band signal of the current channel to calculate the baseline voltage of the current channel.

[0044] For example, the IED detection module 13 can use a Hampel filter to remove transient artifacts (window width is set to 500 ms, threshold is set to 3 times the local standard deviation) on the first frequency band signal, and then calculate the median as the baseline voltage, so that the sampling points with abnormal local amplitude (such as the interference of the spike part in the IED event) can be eliminated. Of course, the first frequency band signal after removing the transient artifact can also be divided into several segments (for example, ten segments), and the median of each segment is calculated, and then the average of the medians of each segment is calculated as the baseline voltage, which is not limited here. After determining the baseline voltage, the median absolute deviation MAD needs to be calculated:

[0045] (1)

[0046] wherein, is a Gaussian distribution correction coefficient, represents taking the median, is the voltage value of the i-th sampling point in the first frequency band signal, represents the baseline voltage.

[0047] Then, the IED detection module 13 can perform peak value detection on the first frequency band signal of the current channel based on the baseline voltage, and determine the candidate nodes that meet the peak value condition, wherein the peak value condition is that the voltage value of the peak point is higher than the sum of the baseline voltage and 3 times MAD, and MAD represents the median absolute deviation. This scheme determines the detection peak value as the index by analyzing the characteristics of the IED event in the stereoelectroencephalogram data, and determines the sampling points that meet the peak value condition as the candidate nodes. Although this embodiment does not select the valley point of the spike wave, the valley point of the spike wave can also be selected, but the determination of the subsequent target time window (the time window for determining whether the ripple event is coupled with the IED event) needs to be adjusted accordingly. The window size needs to be controlled at about 30 ms.

[0048] For each candidate node, the IED detection module 13 can determine a 200 ms interval forward and a 400 ms interval backward based on the candidate node to obtain a 600 ms time window corresponding to the candidate node. Then, the spike wave form verification, slow wave oscillation verification and background suppression verification are performed on the signal in the time window. ​

[0049] For example, based on the peak point corresponding to the candidate node, the first valley point after the peak point in the time window is determined (in most cases, this valley point is the spike valley point of the IED event), and then the first 40% valley value points before and after the valley point are determined. The spike waveform symmetry index is calculated to determine the spike waveform verification result.

[0050] Specifically, the first 40% valley value points before and after the valley point can be determined respectively. The first 40% valley value point before the valley point is the point at which the value decreases from the baseline voltage to 40% of the amplitude difference between the baseline voltage and the valley point based on the baseline voltage. The first 40% valley value point after the valley point is the point at which the value increases by 60% from the valley point to the baseline voltage (i.e., the potential difference between the baseline voltage and the valley point is 60% of the potential difference between the baseline voltage and the valley point). Then the time points of the first 40% valley value point before the valley point, the time point of the valley point, and the time point of the first 40% valley value point after the valley point are determined. The spike waveform symmetry index is calculated according to the following formula:

[0051] (2)

[0052] wherein, is the spike waveform symmetry index, is the time point of the valley point, is the time point of the first 40% valley value point before the valley point, is the time point of the first 40% valley value point after the valley point.

[0053] If , it is determined that the spike waveform verification is passed, otherwise, the spike waveform verification is not passed.

[0054] For example, the IED detection module 13 can determine the first valley point after the peak point in the time window based on the peak point corresponding to the candidate node, and then determine the two valley points after the first valley point. The oscillation decay verification result is determined based on the values of each valley point.

[0055] Specifically, the IED detection module 13 can determine the first valley point after the peak point in the time window (in most cases, this valley point is the spike valley point of the IED event), and then determine two valley points after the first valley point. It is determined whether the voltage values of the valley points located in the rear are greater than the voltage values of the valley points in the front (i.e., V1 is the voltage value of the first valley point after the peak point, V2 is the voltage value of the first valley point after the first valley point after the peak point, and V3 is the voltage value of the second valley point after the first valley point after the peak point). If this condition is met, it is determined that the oscillation decay verification is passed; otherwise, the oscillation decay verification is not passed.

[0056] For example, the IED detection module 13 can determine a peak point before the peak point corresponding to the candidate node from the time window, and determine a background suppression verification region (with the start point of the time window as the start point of the background suppression verification region, and the peak point before the peak point corresponding to the candidate node as the end point of the background suppression verification region) from the time window with the peak point before the peak point corresponding to the candidate node as the end point, and then determine whether there is a sampling point in the background suppression verification region that meets the following conditions:

[0057] (3)

[0058] wherein, V (t) represents the voltage value of any sampling point in the background suppression verification region, V (t) represents the voltage value of any sampling point in the background suppression verification region, is the baseline voltage, is the voltage value of the peak point before the peak point corresponding to the candidate node.

[0059] If there is no sampling point in the background suppression verification region that meets the condition (formula (3)), it is determined that the background suppression verification is passed. Otherwise, it is determined that the background suppression verification is not passed.

[0060] After the verification of the spike waveform, the slow wave oscillation and the background suppression is completed, the IED detection module 13 can determine the marked candidate node based on the verification results (whether the verification is passed) of the verification of the spike waveform, the slow wave oscillation and the background suppression. In this embodiment, if any two of the verification of the spike waveform, the slow wave oscillation and the background suppression are passed, the candidate node is determined as the marked candidate node (because the typical waveform of the IED event generally meets 2-3 conditions, and the atypical and difficult-to-detect waveform, such as the multi-spike slow wave cluster, generally meets two conditions, so as to realize the detection of the IED event).

[0061] After the marked candidate node is determined, the determination of the IED event start point and the IED event end point is required. The IED detection module 13 can determine the first sampling point reaching the baseline voltage as the IED event start point with the peak point corresponding to the marked candidate node. In addition, the IED detection module 13 can take the second valley point after the valley point after the peak point corresponding to the marked candidate node as the start point, search for the first target interval with a voltage value less than the sum of the baseline voltage and 2 times the MAD in 80 ms continuously in the interval of 200 ms, and if the target interval exists, take the start point of the target interval as the IED event end point, and if the target interval does not exist, take the position of 80 ms before the end point of the time window where the marked candidate node is located as the IED event end point.

[0062] After determining the IED event start point and the IED event end point corresponding to each marker candidate node, integration can be performed, i.e., integrating IED events that exist in the intersection of the IED event start point and the IED event end point, so that multiple IED events that actually belong to the same multi-spine slow wave cluster but are identified as different IED events can be integrated into one IED event node, improving the detection accuracy of the IED events. Accordingly, the IED detection module 13 can determine all IED event nodes corresponding to the first frequency band signal of each channel.

[0063] At the same time, the ripple wave detection module 14 can perform sliding window analysis on the second frequency band signal to detect the ripple wave signal and determine the ripple wave event node.

[0064] In this embodiment, for the second frequency band signal of each channel: the ripple wave detection module 14 can perform global analysis on the second frequency band signal of the current channel to calculate the baseline voltage of the current channel. The calculation of the baseline voltage and the calculation of the MAD can refer to the description of the related processing process of the first frequency band signal in the foregoing, which will not be described here.

[0065] Then, the ripple wave detection module 14 can perform sliding window processing on the second frequency band signal of the current channel to obtain a plurality of window data, wherein the sliding window size is 100 ms and the step length is 20 ms. The sliding window size of 100 ms and the step length of 20 ms here are mainly designed to consider the short-term characteristics of physiological ripple waves.

[0066] For each window data: the ripple wave detection module 14 can perform feature extraction on the window data to determine a plurality of feature indexes.

[0067] For example, the ripple wave detection module 14 can perform energy envelope calculation on the window data to obtain the window energy envelope.

[0068] Specifically, the window energy envelope is calculated as follows:

[0069] , (4)

[0070] wherein, is the window energy envelope of the window data, is the number of sampling points of the window data, is the sampling voltage value of the th data in the window data.

[0071] In addition, the ripple wave detection module 14 can perform wave peak number statistics on the window data to obtain the wave peak number of each window data.

[0072] Then, the ripple wave detection module 14 can determine whether there are three or more continuous window data whose window energy envelope is greater than the sum of the baseline voltage of the previous channel and 4 times the MAD, and the number of peaks of each window data is not less than 8.

[0073] If there are such continuous window data not less than 3, the continuous window data are merged as a candidate ripple wave event. Accordingly, the ripple wave detection module 14 can further determine the starting node and the ending node of each candidate ripple wave event.

[0074] For example, for each candidate ripple wave event, the ripple wave detection module 14 can determine, from the first window data of the candidate ripple wave event, the time node at which the voltage amplitude first exceeds the sum of the baseline voltage and 2 times the MAD as the starting node of the candidate ripple wave event, and determine, from the last window data of the candidate ripple wave event, the time node at which the voltage amplitude last exceeds the sum of the baseline voltage and 2 times the MAD as the ending node of the candidate ripple wave event.

[0075] In this way, the ripple wave detection module 14 can obtain a plurality of ripple wave event nodes with determined starting nodes and ending nodes. The second frequency band signal of each channel is processed in this way, and the ripple wave event nodes in the second frequency band signal of each channel can be obtained.

[0076] The decision recognition module 15 needs to determine the ripple wave signal category based on the IED event node and the ripple wave event node, wherein the ripple wave signal category includes physiological ripple wave and pathological signal.

[0077] In this embodiment, the decision recognition module 15 can use a multi-level discrimination method to distinguish physiological ripple wave and pathological signal.

[0078] First, for each channel: the decision recognition module 15 can perform event node matching based on the IED event node and the ripple wave event node to determine that the ripple wave event node with the starting node located in the target time window is a pathological signal, wherein the target time window represents a 50ms time window starting from the peak point corresponding to the IED event node (if the sharp wave valley point is used as the reference, a 30ms time window backward is determined as the target time window).

[0079] For the rest of the ripples event nodes which are not identified as pathological signals, the decision recognition module 15 can perform a fast spectral analysis and energy concentration analysis on the merged window data where the ripples event node is located (after analyzing the frequency domain characteristics of physiological ripples and pathological signals, we found that the spectral peak frequency of physiological ripples is mainly distributed in 70-120Hz, while the spectral peak frequency of pathological signals is mainly distributed in 140-180Hz; and in terms of energy concentration, physiological ripples are generally lower than 0.2, while pathological signals are usually not lower than 0.35, both of which have strong distinguishability, so they are selected), determine the spectral peak frequency and energy concentration, and then determine whether the ripples event node is a pathological signal based on the spectral peak frequency and energy concentration.

[0080] For example, the decision recognition module 15 can perform a fast spectral analysis on the merged window data where the ripples event node is located.

[0081] First, the merged window data can be preprocessed by adding a Hanning window to suppress spectral leakage:

[0082] , (5)

[0083] wherein, is the value of the th sampling point of the windowed merged window data, is the value of the th sampling point of the merged window data, , is the number of sampling points in the merged window data.

[0084] Then, the number of sampling points in the merged window data after adding the Hanning window is expanded to the nearest power of 2 (such as 2000 sampling points to 2048 sampling points) by zero padding (adding zeros at the end) through fast Fourier transform (FFT), i.e. m data ( ), and then the spectral calculation is performed:

[0085] , (6)

[0086] wherein, is the Fourier transform of the merged window data, is the frequency, is the imaginary unit, is the number of sampling points after expansion of the merged window data.

[0087] Accordingly, the decision recognition module 15 can calculate the spectral peak frequency:

[0088] , (7)

[0089] wherein, the spectral peak frequency of the merged window data, the frequency, the Fourier transform of the merged window data.

[0090] And the decision recognition module 15 can perform energy concentration analysis on the merged window data where the ripples event node is located, and calculate the energy concentration:

[0091] , (8)

[0092] wherein, the energy concentration of the merged window data, is ~ the energy integral of the frequency band, is the energy integral of the full frequency band of 70-180 Hz.

[0093] After calculating the spectral peak frequency and energy concentration of the merged window data, the decision recognition module 15 can determine whether the spectral peak frequency of the merged window data where the ripples event node is located is not lower than 140 Hz, and the energy concentration is not lower than 0.35.

[0094] If the spectral peak frequency of the merged window data where the ripples event node is located is not lower than 140 Hz, and the energy concentration is not lower than 0.35, the decision recognition module 15 can determine that the ripples event node is a pathological signal.

[0095] Then, the decision recognition module 15 can determine whether the spectral peak frequency of the merged window data where the ripples event node is located is not more than 120 Hz, and the energy concentration is not more than 0.2.

[0096] If the spectral peak frequency of the merged window data where the ripples event node is located is not more than 120 Hz, and the energy concentration is not more than 0.2, the decision recognition module 15 can determine that the ripples event node is a physiological ripple.

[0097] If the spectral peak frequency of the merged window data where the ripples event node is located is within (120, 140), and the energy concentration is within (0.2, 0.35), such ripples event node is relatively rare, and can be classified as physiological ripples and pathological signals according to a bottom-up processing method: calculating the probability that the ripples event node belongs to a pathological signal based on the spectral peak frequency and the energy concentration, and determining whether the ripples event node is a pathological signal.

[0098] For example, the probability that the ripples event node belongs to a pathological signal is calculated according to the deviation of the spectral peak frequency and the energy concentration from the judgment standard:

[0099] , (9)​

[0100] wherein, is the probability that the ripples event node belongs to pathological signal, and are weight indexes, and For example, the present embodiment takes and each being 0.5 as an example, is the spectral peak frequency of the ripples event node, is the energy concentration degree of the ripples event node.

[0101] If the probability that the ripples event node belongs to pathological signal is not less than 0.5, it is determined that the ripples event node belongs to pathological signal, and accordingly the classification and identification of the ripples event are completed.

[0102] For each channel: based on the IED event node and the ripples event node, event node matching is performed, and it is determined that the ripples event node whose starting node is located in the target time window is a pathological signal, wherein the target time window represents a 50 ms time window with the peak point corresponding to the IED event node as the starting point; for the ripples event node in each channel that is not determined to be a pathological signal: the merged window data in which the ripples event node is located is subjected to fast spectral analysis and energy concentration degree analysis, the spectral peak frequency and the energy concentration degree are determined, and then based on the spectral peak frequency and the energy concentration degree, it is determined whether the ripples event node is a pathological signal.

[0103] In summary, the embodiment of the present application provides a ripples signal detection and recognition system 10 based on IED events and time-frequency characteristics. The signal acquisition module 11 acquires the EEG data of the target object (including the channel signals of each hippocampal subregion, and the hippocampal subregion includes CA1-CA4, Subiculum, Presubiculum, and DG). The pre-processing module 12 performs power frequency notch filtering and band-pass filtering on each channel signal of the EEG data. The power frequency notch filtering removes 50Hz power frequency interference, and the band-pass filtering retains the first frequency band signal of the 25-80Hz frequency band and the second frequency band signal of the 70-180Hz frequency band, respectively. The IED detection module 13 performs IED signal detection on the first frequency band signal to determine the IED event node. The ripples detection module 14 performs sliding window analysis on the second frequency band signal to perform ripples signal detection and determine the ripples event node. The decision recognition module 15 determines the ripples signal category (physiological ripples or pathological signal) based on the IED event node and the ripples event node. The system acquires the hippocampal subregion through the signal acquisition module 11, separates the double frequency bands (25-80Hz and 70-180Hz) through the pre-processing module 12, and constructs a time-space coupled detection basis. The IED event is used as the time-space anchor point of the pathological signal (the specific construction target time window is used as the judgment basis), which assists the recognition and classification of the pathological signal. At the same time, for the ripples event not coupled with IED, the essential difference between physiological ripples and pathological signal in frequency spectrum is analyzed, and the double-index discrimination of spectral peak frequency and energy concentration degree is designed to essentially distinguish the pathological signal and physiological ripples. The pathological signal appearing independently can be effectively recognized, the accuracy of ripples category division is improved, and the probability of misclassification is reduced.

[0104] In the identification of IED events, by analyzing the waveform form at the occurrence of IED, the detection mechanism of IED event is designed, and a variety of discrimination standards (based on the peak point corresponding to the candidate node, the first valley point is determined after the peak point in the time window, the first 40% valley value point before and after the valley point is determined, the symmetry index of the spike waveform form is calculated, and the verification result of the spike waveform form is determined; based on the peak point corresponding to the candidate node, the first valley point is determined after the peak point in the time window, and then the first two valley points after the first valley point are determined, and the verification result of the oscillation attenuation is determined based on the value of each valley point; based on the peak point corresponding to the candidate node, the peak point before the peak point in the time window is determined, and the background suppression verification area is determined with the previous peak point as the end point, and the background suppression verification result is determined) are superimposed, which can effectively deal with various typical forms of IED events (including spike wave, sharp wave, spike slow complex wave, etc.), improve the accuracy of IED event identification, and based on the coupling window characteristics of pathological signals and IED events in the previous research results, the coupling standard of IED event node as pathological signal (50ms time window with the peak point corresponding to the IED event node as the starting point) is determined, which improves the identification accuracy of pathological signals coupled with IED events.

[0105] In the detection process of the ripple wave, the baseline voltage is calculated by globally analyzing the second frequency band signal, and then the sliding window processing is performed (the sliding window size is 100ms, and the step is 20ms, which can consider the short-term characteristics of physiological ripple wave, improve the detection accuracy of ripple wave event, and reduce the missed detection rate), and a plurality of window data are obtained. For each window data: the window energy envelope is obtained by calculating the energy envelope of the window data; the number of wave peaks of the window data is counted; it is judged whether there are three or more continuous window data whose window energy envelope is greater than the sum of the baseline voltage of the previous channel and 4 times the MAD (median absolute deviation), and the number of wave peaks of each window data is not less than 8, if there are, the continuous window data are merged as a candidate ripple wave event. And the time node at which the voltage amplitude first exceeds the sum of the baseline voltage and 2 times the MAD is determined as the starting node of the candidate ripple wave event from the first window data of the candidate ripple wave event; the time node at which the voltage amplitude last exceeds the sum of the baseline voltage and 2 times the MAD is determined as the termination node of the candidate ripple wave event from the last window data of the candidate ripple wave event. This method not only can effectively detect the ripple wave event, but also can relatively accurately determine the starting range of the ripple wave event, so as to improve the discrimination accuracy and reduce the false positive probability (i.e. physiological ripple wave is misidentified as pathological signal) in the subsequent IED event coupling discrimination process.

[0106] The above merely provides an example of the present application, and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A ripple wave signal detection and recognition system based on IED events and time-frequency characteristics, characterized in that, include: The signal acquisition module is used to acquire the EEG data of the target object. The EEG data contains at least the channel signals of each hippocampal subregion, which includes CA1-CA4, Subiculum, Presubiculum, and DG. CA1-CA4 represent hippocampal horn regions 1-4, Subiculum represents the inferior horn, Presubiculum represents the anterior inferior horn, and DG represents the dentate gyrus. The preprocessing module is used to preprocess the signals of each channel of the EEG data. The preprocessing includes power frequency notch filtering and bandpass filtering. Power frequency notch filtering removes 50Hz power frequency interference, and bandpass filtering retains the 25-80Hz frequency band and the 70-180Hz frequency band respectively. The 25-80Hz frequency band is the first frequency band signal, and the 70-180Hz frequency band is the second frequency band signal. The IED detection module is used to detect IED signals in the first frequency band signal and determine the IED event node, where IED represents interictal epileptiform discharge. The ripple wave detection module is used to perform sliding window analysis on the second frequency band signal, detect ripple wave signals, and determine ripple wave event nodes. The decision recognition module is used to determine the ripple wave signal category based on the IED event node and the ripple wave event node. The ripple wave signal category includes physiological ripple waves and pathological signals. The decision recognition module is specifically used for: For each channel: Event node matching is performed based on IED event nodes and ripple wave event nodes. Ripple wave event nodes whose starting node is located within the target time window are identified as pathological signals. The target time window is a 50ms time window starting from the peak point corresponding to the IED event node. For ripple wave event nodes in each channel that are not identified as pathological signals: perform rapid spectrum analysis and energy concentration analysis on the merged window data where the ripple wave event node is located to determine the peak frequency and energy concentration, and then determine whether the ripple wave event node is a pathological signal based on the peak frequency and energy concentration. The decision recognition module is specifically used for: Perform fast spectral analysis on the merged window data containing the ripple wave event nodes to calculate the spectral peak frequencies: ， in, To merge the spectral peak frequencies of the window data, For frequency, Fourier transform for merging window data; Energy concentration analysis is performed on the merged window data containing the ripple wave event nodes, and the energy concentration is calculated: ， in, To optimize the energy concentration of merged window data, for ~ Energy integral of frequency band, The energy integral is the energy across the entire frequency band from 70 to 180 Hz.

2. The ripple wave signal detection and recognition system based on IED events and time-frequency characteristics according to claim 1, characterized in that, The IED detection module is specifically used for: For the first frequency band signal of each channel: Perform a global analysis on the first frequency band signal of the current channel to calculate the baseline voltage of the current channel; Using the baseline voltage as a reference, peak detection is performed on the first frequency band signal of the current channel, and candidate nodes that meet the peak condition are identified. The peak condition is: higher than the sum of the baseline voltage and 3 times MAD, where MAD represents the median absolute deviation. For each candidate node, a 200ms interval is determined forward and a 400ms interval is determined backward, resulting in a 600ms time window for the candidate node. Then, spike morphology verification, slow wave oscillation verification, and background suppression verification are performed on the signal within the time window. Based on the verification results of spike morphology verification, slow wave oscillation verification, and background suppression verification, the marked candidate nodes that meet the requirements are determined. Based on the marked candidate nodes, the IED event start point and IED event end point are determined.

3. The ripple wave signal detection and recognition system based on IED events and time-frequency characteristics according to claim 2, characterized in that, The IED detection module is specifically used for: Based on the peak points corresponding to the candidate nodes, the first valley point after the peak point is determined within the time window, the first 40% valley amplitude point before and after the valley point is determined, the spike morphology symmetry index is calculated, and the spike morphology verification result is determined. Based on the peak points corresponding to the candidate nodes, the first valley point after the peak point is determined within the time window, and then the two valley points after the first valley point are determined. The oscillation decay verification result is determined based on the value of each valley point. Based on the peak point corresponding to the candidate node, the previous peak point is determined within the time window, and the background suppression verification region is determined within the time window with the previous peak point as the endpoint, thus determining the background suppression verification result.

4. The ripple wave signal detection and recognition system based on IED events and time-frequency characteristics according to claim 1, characterized in that, The ripple wave detection module is specifically used for: For the second frequency band signal of each channel: Perform a global analysis on the second frequency band signal of the current channel to calculate the baseline voltage of the current channel; The second frequency band signal of the current channel is processed by sliding window to obtain several window data, wherein the sliding window size is 100ms and the step size is 20ms; For each window of data: feature extraction is performed on the window data to determine several feature indices. Based on the feature indices, candidate ripple events are determined. Then, the start and end nodes of each candidate ripple event are determined to obtain several ripple event nodes.

5. The ripple wave signal detection and recognition system based on IED events and time-frequency characteristics according to claim 4, characterized in that, The ripple wave detection module is specifically used for: The energy envelope of the window is obtained by performing energy envelope calculation on the window data; Count the number of peaks in the window data; Determine if there are three or more consecutive window energy envelopes that are greater than the sum of the baseline voltage of the front channel and four times the MAD, and if each window data has at least eight peaks. If so, merge the consecutive window data as a candidate ripple wave event, where MAD represents the median absolute deviation.

6. The ripple wave signal detection and recognition system based on IED events and time-frequency characteristics according to claim 5, characterized in that, The ripple wave detection module is specifically used for: The window energy envelope is calculated using the following method: ， in, The window energy envelope of the window data. The number of sampling points for the window data. For the first data in the window The sampled voltage values ​​of each data point.

7. The ripple wave signal detection and recognition system based on IED events and time-frequency characteristics according to claim 5, characterized in that, The ripple wave detection module is specifically used for: For each candidate ripple wave event: From the first window of data of the candidate ripple wave event, the time point at which the voltage amplitude first exceeds the sum of the baseline voltage and twice the MAD is determined as the starting point of the candidate ripple wave event; From the last window of data of the candidate ripple wave event, the time point at which the voltage amplitude last exceeds the sum of the baseline voltage and twice the MAD is determined as the termination point of the candidate ripple wave event.

8. The ripple wave signal detection and recognition system based on IED events and time-frequency characteristics according to claim 3, characterized in that, The decision recognition module is specifically used for: If the spectral peak frequency of the merged window data where the ripple wave event node is located is not lower than 140Hz and the energy concentration is not lower than 0.35, the ripple wave event node is determined to be a pathological signal. If the spectral peak frequency of the merged window data where the ripple wave event node is located does not exceed 120Hz and the energy concentration does not exceed 0.2, the ripple wave event node is determined to be a physiological ripple wave. If the spectral peak frequency of the merged window data where the ripple wave event node is located is within (120, 140) and the energy concentration is within (0.2, 0.35), calculate the probability that the ripple wave event node belongs to a pathological signal based on the spectral peak frequency and energy concentration, and determine whether the ripple wave event node is a pathological signal.

Images