Method, device and storage medium for long-term monitoring of epileptic seizures

Abstract

The present application discloses a method, device and storage medium for long-term monitoring of epileptic seizures, and selects a single EEG channel and four EEG features. The present application filters and segmentally pre-processes the EEG signals of all channels pre-measured by the patient, uses the power spectral density function (PSD) parameterization technology for feature extraction, characterizes the non-periodic component with two parameters, the offset and the exponent, selects the two largest powers and their corresponding center frequencies in the periodic component, and the original PSD corresponding to the center frequencies of the two periodic components is recorded as the total feature power; selects the four key features most relevant to the seizure event; selects the most representative channel according to the statistical significance P value of the four key parameterized features during the seizure period and the inter-seizure period, and performs data enhancement on the EEG data of the channel; finally, performs automatic detection of epileptic seizures based on the support vector machine classifier. The present application does not require pre-specified frequency sub-bands, and can adaptively select representative channels for specific patients. It has good tolerance for individual differences, can achieve consistent performance for each patient, and has strong robustness.

Description

Technical Field

[0001] The present application relates to the field of long-term monitoring of epileptic seizures, and in particular to a method, device and storage medium for long-term monitoring of epileptic seizures. Background Art

[0002] As an emerging development, wearable, minimal-channel automated epilepsy monitoring systems still face challenges in meeting clinical application requirements, including relatively low sensitivity (i.e., high missed detection rates). This may be primarily due to two factors. First, most current epileptic seizure detection algorithms are still based on typical frequency band analysis (frequency bands are primarily categorized as δ: 1-4 Hz, θ: 4-8 Hz, α: 8-12 Hz, and β: 12-30 Hz). For example, the 13 time-domain and frequency-domain features extracted by Swinnen et al. are all obtained within or between fixed frequency bands; Janjarasjitt identified the most effective frequency subbands for epileptic seizure detection using wavelet-based features; while the Hilbert-Huang transform used by Lu Yanan et al. can adaptively extract features, some features are still extracted based on frequency subbands; and Romaine et al. calculated envelope-based phase synchronization features within the fixed β-12-30 Hz frequency band. Second, the extracted features fail to effectively represent key information in ictal and interictal EEG signals. Researchers in computer-assisted epilepsy medicine typically rely solely on traditional one-dimensional time-series signal processing techniques, lacking prior knowledge of epileptic seizures. This results in models that tend to be generalized, contradicting the original "task-driven" approach. For example, Romanine et al. used a fixed beta frequency band for seizure detection, a subband that has little correlation with the frequency of ictal EEG signals.

[0003] Therefore, the above-mentioned technical problems existing in related technologies need to be solved urgently. Summary of the Invention

[0004] The present application aims to solve one of the technical problems in the related art. To this end, the embodiments of the present application provide a method, device and storage medium for long-term monitoring of epileptic seizures, which can monitor the duration of epileptic seizures.

[0005] According to one aspect of the embodiments of the present application, a method for long-term monitoring of epileptic seizures is provided, the method comprising:

[0006] Obtain EEG data of all channels;

[0007] The feature extraction is performed using segmented EEG PSD parameterization, where the non-periodic component is characterized by two parameters: offset and exponent. The periodic component is characterized by the two maximum powers PW1 and PW2 and their corresponding center frequencies CF1 and CF2. The characteristic total powers TPW1 and TPW2 correspond to the original PSD values corresponding to the center frequencies CF1 and CF2, respectively.

[0008] Select several key parameterized features that are most relevant to seizure events;

[0009] Selecting the most representative channels based on the statistically significant P values of the key parameterized features during ictal and interictal periods;

[0010] Automatic epileptic seizure detection is performed based on the support vector machine classifier according to the EEG signal features of the most representative channels.

[0011] In one embodiment, the method further comprises:

[0012] The data of all channels are filtered and EEG signals are segmented and preprocessed.

[0013] In one embodiment, data from all channels are filtered and EEG signal segmentation preprocessing is performed, including:

[0014] The EEG data were filtered using a high-pass filter and a low-pass filter with zero phase shift;

[0015] The preictal and postictal periods were defined as the 10 minutes before and after the ictal period, respectively, and the part excluding the preictal, postictal, and ictal periods was defined as the interictal period;

[0016] Sliding 5-second time windows were used to segment ictal and interictal EEG signals without overlap, and the DC component of the EEG signals was removed;

[0017] The cutoff frequency of the high-pass filter is -3dB, 0.5Hz, and the cutoff frequency of the low-pass filter is -3dB, 45Hz.

[0018] In one embodiment, after selecting the most representative channel based on the statistically significant P values of the key parameterized features during ictal and interictal periods, the method further comprises:

[0019] Data enhancement is performed on the EEG data of the most representative channel.

[0020] In one embodiment, the mathematical formula for calculating the PSD function is:

[0021] PSD=L+G

[0022] L=b-log10(k+F χ )

[0023]

[0024]

[0025] Where PSD is the power spectral density (PSD) function, L is the non-periodic component, G is the periodic component, b is the offset, χ is the exponent, F is the input frequency vector, ranging from 0.5 to 30 Hz, with an interval of 0.5 Hz, k is the "inflection point" parameter used to control the bending of the non-periodic component, which is set to 0 in this application, N is the number of Gaussian function combinations, and the maximum value is set to 8 in this application, a i is the peak power, expressed as log10(Power), c i and w i is the center frequency and standard deviation of the Gaussian function.

[0026] In one embodiment, several key parameterized features with the highest correlation with the seizure event are selected, and the most representative channels are selected based on the statistical significance P values of the key parameterized features during the seizure period and the interictal period, including:

[0027] Ranking based on the statistical significance P value of key parameters in the interictal and ictal periods;

[0028] Select the most representative channels based on the ranking;

[0029] The key parameterized features include four items: offset, index, TPW1 and TPW2.

[0030] In one embodiment, automatic epileptic seizure detection is performed based on a support vector machine classifier according to the above four features of the most representative channels, including the following data augmentation method:

[0031] Determine the 5 nearest neighbors of the sample and randomly select N nearest neighbors from them;

[0032] Each sample is subtracted from the selected neighboring sample and then multiplied by a random factor between 0 and 1 to obtain a new synthetic sample;

[0033] The amplified data were input into the support vector machine to classify the ictal and interictal states;

[0034] Here, N is adaptively set to an integer ratio between the number of samples in the majority class and the minority class.

[0035] According to one aspect of the present application, a device for long-term monitoring of epileptic seizures is provided, the device comprising:

[0036] The first module is used to obtain data from all channels;

[0037] The second module is used to extract features using PSD parameterization, where the non-periodic component is characterized by two parameters: offset and exponent, and the periodic component is characterized by the two maximum powers PW1 and PW2 and their corresponding center frequencies CF1 and CF2. The characteristic total powers TPW1 and TPW2 correspond to the original PSD values corresponding to the center frequencies CF1 and CF2, respectively.

[0038] The third module is used to select several key parameterized features that are most relevant to the seizure event;

[0039] A fourth module is configured to select the most representative channel based on the statistically significant P values of the key parameterized features during the ictal and interictal periods;

[0040] The fifth module is used to automatically detect epileptic seizures based on the support vector machine classifier according to the four features of the most representative channel EEG (non-periodic component using offset and index, feature total power TPW1 and TPW2).

[0041] According to one aspect of the present application, a device for long-term monitoring of epileptic seizures is provided, the device comprising:

[0042] a single-channel EEG data acquisition device;

[0043] at least one processor;

[0044] At least one memory, the memory being used to store at least one program and EEG data collected by the above-mentioned device;

[0045] When at least one of the programs is executed by at least one of the processors, the long-term monitoring method for epileptic seizures as described in the previous embodiment is implemented.

[0046] According to one aspect of an embodiment of the present application, a storage medium is provided, wherein the storage medium stores a program executable by a processor, and when the program executable by the processor is executed by the processor, the method for long-term monitoring of epileptic seizures as described in the previous embodiment is implemented.

[0047] The beneficial effects of the long-term epileptic seizure monitoring method, device, and storage medium provided by the embodiments of the present application are as follows: the method of the present application includes: acquiring data from all channels; using PSD parameterization for feature extraction, wherein the non-periodic component is characterized by two parameters, offset and exponent, and the periodic component is characterized by two maximum powers PW1 and PW2 and their corresponding center frequencies CF1 and CF2, and the characteristic total powers TPW1 and TPW2 correspond to the original PSD values corresponding to the center frequencies CF1 and CF2, respectively; selecting several key parameterized features with the highest correlation with the seizure event; selecting the most representative channel based on the statistical significance P value of the key parameterized features during the seizure period and the interictal period; and automatically detecting epileptic seizures based on the most representative channel using a support vector machine classifier. The present application can adapt to the EEG power spectrum of different patients. For example, Swinnen et al. extracted 13 features in the time and frequency domains, all of which were extracted within or between pre-specified fixed frequency bands; Janjarasjitt used eight wavelet-based features to identify the most commonly used frequency subbands in epileptic EEG signals; Lu Yanan et al. extracted 12 features based on the Hilbert-Huang transform and Kraskov entropy, some of which were adaptive and some were not; and Romanie calculated envelope-based phase synchronization features in the β-12-30 Hz frequency band, where the β band has little frequency correlation with the ictal EEG signal. Because the present application does not require pre-specified frequency subbands for extraction and can adaptively select representative channels for specific patients, it is very tolerant of individual differences in the power spectrum. Therefore, the present application can achieve consistent performance for each patient, meaning that the monitoring model has considerable robustness.

[0048] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become obvious from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0050] Figure 1 A flowchart of a method for long-term monitoring of epileptic seizures provided in an embodiment of the present application;

[0051] Figure 2 A schematic diagram of the working process of the long-term monitoring method for epileptic seizures provided in an embodiment of the present application;

[0052] Figure 3Schematic diagram of EEG signals during the ictal and interictal periods provided in the embodiments of the present application;

[0053] Figure 4 Schematic diagram of typical ictal and interictal EEG signal segments provided in the embodiments of the present application;

[0054] Figure 5 A schematic diagram of a long-term epileptic seizure monitoring device provided in an embodiment of the present application;

[0055] Figure 6 A schematic diagram of another long-term epileptic seizure monitoring device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0057] The terms "first," "second," "third," and "fourth," etc., in the specification, claims, and drawings of this application are used to distinguish between different objects, not to describe a particular order. In addition, the terms "including," "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements, but may optionally include steps or elements not listed, or may optionally include other steps or elements inherent to the process, method, product, or apparatus.

[0058] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0059] Long-term, automated home monitoring systems for epileptic seizures require two fundamental qualities: effectiveness and ease of use. The former requires the monitoring system to capture the EEG signal characteristics of epileptic seizures and accurately and efficiently identify the patient's epileptic seizure state; the latter requires the long-term monitoring system to be easy to operate, compact, low-cost, and portable. Traditional automated epileptic seizure monitoring technologies rely on long-term multi-channel EEG signals and EEG feature extraction based on classical frequency band decomposition. To achieve effective, economical, and unobtrusive long-term monitoring, the system must be lightweight, low-power, and low-complexity. Therefore, researchers have proposed single-channel or dual-channel epileptic seizure monitoring systems. Swinnen et al. proposed a dual-channel, wearable, behind-the-ear EEG monitoring system. They extracted three time-domain features and ten frequency-domain features in different frequency bands for typical absence seizure detection, achieving a sensitivity and specificity of 98.3% and 99%, respectively. Janjarasjit et al. used a single-channel EEG signal to extract eight wavelet-based features to identify the most common frequency subbands in epileptic EEG signals. In epileptic seizure detection, its detection accuracy reaches 97%, but the sensitivity is low, with an average of 73%. The low detection sensitivity is caused by the imbalance of the data set for training the classifier; Lu Yanan et al. used 12 features in the Hilbert-Huang transform and Kraskov entropy as the classification of EEG segments during epileptic seizures and interictal periods for EEG signal detection of single-channel epileptic seizures, and their classification accuracy and sensitivity were 72% and 75%, respectively; Romaine et al. used single-channel EEG signals to calculate the envelope phase synchronization in the β~12-30Hz frequency band for epileptic seizure detection, achieving 83% sensitivity and 75% specificity.

[0060] As an emerging development, wearable, minimal-channel automated epilepsy monitoring systems still face challenges in meeting clinical application requirements, including relatively low sensitivity (i.e., high missed detection rates). This may be primarily due to two factors. First, most current epileptic seizure detection algorithms are still based on typical frequency band analysis (frequency bands are primarily categorized as δ: 1-4 Hz, θ: 4-8 Hz, α: 8-12 Hz, and β: 12-30 Hz). For example, the 13 time-domain and frequency-domain features extracted by Swinnen et al. are all obtained within or between fixed frequency bands; Janjarasjitt identified the most effective frequency subbands for epileptic seizure detection using wavelet-based features; while the Hilbert-Huang transform used by Lu Yanan et al. can adaptively extract features, some features are still extracted based on frequency subbands; and Romaine et al. calculated envelope-based phase synchronization features within the fixed β-12-30 Hz frequency band. Second, the extracted features fail to effectively represent key information in ictal and interictal EEG signals. Researchers in computer-assisted epilepsy medicine typically rely solely on traditional one-dimensional time-series signal processing techniques, lacking prior knowledge of epileptic seizures. This results in models that tend to be generalized, contradicting the original "task-driven" approach. For example, Romanine et al. used a fixed beta frequency band for seizure detection, a subband that has little correlation with the frequency of ictal EEG signals.

[0061] In summary, how to improve the sensitivity of a single-channel long-term monitoring system for epileptic seizures is still an open research problem. This problem requires that the designed algorithm can better characterize the key characteristics of epileptic EEG signals. At present, epileptologists have accumulated rich experience and knowledge in identifying EEG signals of states related to epileptic seizures, and can provide quite authoritative guidance on the oscillation amplitude and related frequencies of the signal. On the other hand, the latest developments in engineering have shown that the EEG power spectrum can be effectively decomposed into non-periodic and periodic components, that is, it can be parameterized into a small number of features related to amplitude and frequency. Therefore, the present application aims to extract key features related to epileptic seizure EEG signals, which match the epilepsyologists' description of the mechanism of epileptic seizures, and then use the extracted features to realize a long-term monitoring system for epileptic seizures based on single-channel EEG.

[0062] Figure 1 This is a flow chart of the long-term monitoring method for epileptic seizures provided in an embodiment of the present application. Figure 2 A schematic diagram of the working process of the long-term monitoring method for epileptic seizures provided in an embodiment of the present application is shown in FIG. Figure 1 and Figure 2 As shown, the long-term monitoring method for epileptic seizures proposed in this application includes:

[0063] S101. Obtain EEG data of all channels.

[0064] S102. Use segmented EEG PSD parameterization for feature extraction.

[0065] The non-periodic component is characterized by two parameters: offset and index. The periodic component is characterized by the two maximum powers PW1 and PW2 and their corresponding center frequencies CF1 and CF2. The characteristic total powers TPW1 and TPW2 correspond to the original PSD values corresponding to the center frequencies CF1 and CF2, respectively.

[0066] S103. Select several key parameterized features that are most correlated with the seizure event.

[0067] S104 , selecting the most representative channel according to the statistically significant P value of the key parameterized features during the attack period and the inter-attack period.

[0068] Optionally, after selecting the most representative channel based on the statistically significant P-values of the key parameterized features during ictal and interictal periods, this embodiment further includes performing data augmentation on the EEG data of the most representative channel. Because the CHB-MIT database suffers from severe data imbalance, with the ratio of epileptic seizure time to non-epileptic seizure time for most patients less than 4%, the synthetic minority oversampling technique (SMOTE) is used for data augmentation to alleviate the problem of SVM bias toward the majority class (interictal period) during classification. The SMOTE algorithm is implemented iteratively. For each minority class sample, its k nearest neighbors are first determined, from which N nearest neighbors are randomly selected. Each sample is subtracted from the selected nearest neighbor sample and multiplied by a factor to obtain a new synthetic sample, where the factor ranges from (0, 1). In this scheme, k is set to 5, and N is adaptively set to the integer ratio between the number of majority and minority class samples. If N is greater than k, then N = k. Finally, the augmented data is input into a support vector machine for classification of ictal and interictal states. The classifier is set to a second-order polynomial kernel function. Five-fold cross-validation is used for training and validation.

[0069] S105. Automatically detect epileptic seizures based on the support vector machine classifier according to the EEG signal features of the most representative channel.

[0070] It should be noted that the data of this application is derived from EEG signal recordings, and the system of this application needs to be based on multi-channel video EEG and epileptic seizure markers of each patient during hospitalization. In order to verify the feasibility of the invention scheme, this application uses a general public database (CHB-MIT scalp EEG database) for testing. The database contains 24 cases of EEG data from 23 epilepsy patients (the 1st and 21st EEG data are from the same patient, recorded one and a half years apart). Most of the case 21 data contain 18 or 23 channels, and a few are 24-26 channels. The data recording follows the EEG electrode position of the international 10-20 system. The sampling rate is 256Hz and the resolution is 16bit. This technology performs parametric analysis on the EEG data of all channels of all cases, and selects a channel to perform the epileptic seizure detection task based on the parametric features most relevant to the seizure event. Therefore, each case data set has 21 channels to be analyzed: FP1-F7, F7-T7, T7-P7, P7-O1, FP1-F3, F3-C3, C3-P3, P3-O1, FP2-F4, F4-C4, C4-P4, P4-O2, FP2-F8, F8-T8, T8-P8, P8-O2, FZ-CZ, CZ-PZ, T7-FT9, FT9-FT10, FT10-T8. Because channels T7-P7 and P7-T7 are recorded from the same pair of electrodes with opposite polarity and therefore provide the same PSD, only one of these channels can be selected for analysis.

[0071] Optionally, this embodiment further includes filtering the data of all channels and pre-processing the EEG signals in segments. Filtering the data of all channels and pre-processing the EEG signals in segments specifically includes: filtering the EEG data using a high-pass filter and a low-pass filter with zero phase shift; defining the pre-ictal and post-ictal periods as the 10 minutes before and after the ictal period, respectively, and defining the period excluding the pre-ictal, post-ictal, and ictal periods as the interictal period; sliding a 5-second time window to segment the ictal and interictal EEG signals without overlap, and removing the DC component of the EEG signals; wherein the cutoff frequency of the high-pass filter is -3dB, 0.5Hz, and the cutoff frequency of the low-pass filter is -3dB, 45Hz.

[0072] Specifically, the data of all channels are filtered and the EEG signal segmentation preprocessing includes: using a high-pass filter with zero phase shift (cut-off frequency: -3dB, 0.5Hz) and a low-pass filter (cut-off frequency: -3dB, 45Hz) to filter the EEG data. The CHB-MIT database provides annotations and calibrations of epileptic seizure events by clinical experts, and this scheme divides the epileptic EEG data into the ictal period and the interictal period based on this. According to the literature, the interictal data can be further divided into the pre-ictal period and the post-ictal period, but the specific duration is not clearly stated. In addition, the interictal period is usually much longer than the ictal period, causing the problem of data imbalance in machine learning. Therefore, this application refers to the relevant literature to define the pre-ictal period and the post-ictal period as 10 minutes before and after the ictal period, and the remaining part is defined as the interictal period. Then, a 5-second time window is slid to segment the ictal and interictal EEG signals without overlapping, and remove their DC components. This application believes that the selection of a 5-second time window is a compromise between balancing the effective capture of epileptic EEG signal characteristics and the assumption of signal stationarity.

[0073] After preprocessing the EEG signal, power spectrum estimation and parameter analysis are required.

[0074] The PSD of the epileptic EEG segments was estimated using the Welch method with a window size of 2 seconds (512 data points) and an overlap of 50%. This PSD estimation algorithm gave a frequency resolution of 0.5 Hz (without zero padding).

[0075] Research has shown that a notable characteristic of EEG signals is that their PSD follows a 1 / f-like distribution. Recent developments in PSD analysis have emphasized the value of decomposing the PSD into aperiodic and periodic components. The aperiodic component refers to the overall 1 / f profile of the PSD, while the periodic component is the bump (narrowband power peak) superimposed on the aperiodic component. Therefore, the periodic component is the difference between the total PSD and its aperiodic component. The aperiodic and periodic components are modeled as Lorentzian and Multi-Gaussian functions, respectively. The Lorentzian function formula for the aperiodic component is:

[0076] L=b-log10(k+F χ )

[0077] Where b is the offset, χ is the exponent, F is the input frequency vector, ranging from 0.5 to 30 Hz, with an interval of 0.5 Hz, and k is the "inflection point" parameter used to control the bending of the non-periodic component, which is set to 0 in this application.

[0078] The periodic component is fitted by multiple Gaussian functions, which is expressed as a combination of N Gaussian functions, where N depends on the number of peaks to be fitted. The maximum value in this application is set to 8, as shown in the formula:

[0079]

[0080] Each Gaussian function is a fitting curve of the periodic component in the PSD, as shown in the formula:

[0081]

[0082] Among them, a i is the peak power, expressed as log10(Power), c i and w i is the center frequency and the standard deviation of the Gaussian function; F is the input frequency vector, ranging from 0.5 to 30 Hz, with an interval of 0.5 Hz.

[0083] Therefore, the total PSD function is finally parameterized as the sum of the non-periodic component L and the periodic component G, as shown in the formula:

[0084] PSD=L+G

[0085] Since most of the EEG signal energy is in the range of 0.5 to 30 Hz and the frequency of epileptic discharge waveforms is relatively low, this scheme parameterizes the PSD within the frequency range of 0.5 to 30 Hz. The process and typical results of the parameterization analysis are shown. Figure 3 and Figure 4 As shown, Figure 3 A 1-hour EEG recording of channels CZ-PZ from case 11 is shown, with EEG signal segmentation performed over a sliding 5-second time window, where Figure 4 Typical interictal (left) and ictal (right) EEG signal segments are shown. Based on this, the present application performed the above process on the 21 channels of ictal and interictal EEG data for each case, analyzed the relationship between the parameters and epileptic seizure events, selected the effective parameters, and tested their effectiveness in epileptic seizure detection.

[0086] This application also includes statistical analysis of PSD parameters, testing the correlation between such parameters and the attack period and the interictal period on a channel-by-channel and case-by-case basis. The statistical analysis specifically includes: first, testing the normality of the parameter distribution. If the parameter follows a normal distribution, the mean and variance of the parameter are calculated; otherwise, the median and interquartile range are calculated. Secondly, for parameters that follow a normal distribution, an independent sample t-test is performed; otherwise, the Kolmogorov-Smirnov test is used. If the two-sided P value is less than 0.05, it is considered statistically significant.

[0087] In addition, in steps S103 and S104, several key parameterized features with the highest correlation with the seizure event are selected, and the most representative channels are selected based on the statistically significant P values of the key parameterized features during the ictal and interictal periods. This includes: ranking the key parameters based on their statistically significant P values during the interictal and ictal periods; selecting the key parameterized features of the most representative channels based on the ranking; the key parameterized features include four items: offset, index, TPW1, and TPW2. As described in Section 4, based on the results of a systematic statistical analysis of all relevant candidate parameters, this scheme ranks the P values of the four key parameterized features (offset, index, TPW1, and TPW2) and ultimately selects an EEG channel for each patient, which represents the patient's interictal and ictal states. The specific ranking process is as follows: Rank each parameter based on its statistically significant P value during the interictal and ictal periods—the channel with the smallest P value is ranked first, and the top eight channels for each of the four parameters are selected; if four parameters simultaneously enter the top eight channels, these channels are selected first for further ranking:

[0088] a) Score the channels according to the ranking of each parameter (Roff, Rexp, RTPW1, RTPW2). For example, if a channel x ranks second in the offset, then Roff_x = 2.

[0089] b) Calculate the total score Rtotal_x of channel x in terms of the ranking of the four parameters;

[0090] c) Sort Rtotal in ascending order to obtain the final channel ranking RCH;

[0091] d) Select the channel ranked first, i.e., the one with the smallest Rtotal, as the representative EEG data of the patient case.

[0092] If no channel has all four parameters in the top 8 at the same time, the channels with three parameters in the top 8 will be ranked.

[0093] Specifically, automatic seizure detection is performed using a support vector machine classifier based on key parameterized features of the most representative channels. This approach involves the following data augmentation: A sample's five nearest neighbors are determined, and N nearest neighbors are randomly selected from these. Each sample is subtracted from the selected nearest neighbors and multiplied by a factor to create a new synthetic sample. The augmented data is then fed into a support vector machine for classification into ictal and interictal states, where N is adaptively set to an integer ratio between the number of samples in the majority and minority classes. To evaluate the parametric features derived from the PSD parameterization technique, this approach uses EEG data from representative channels to extract four key parameters (offset, exponent, TPW1, and TPW2) for automatic seizure detection using support vector machine (SVM) classification.

[0094] Because the CHB-MIT database suffers from severe data imbalance, with the ratio of epileptic seizure time to non-epileptic seizure time in most patients less than 4%, the synthetic minority oversampling technique (SMOTE) was used for data augmentation to mitigate the bias of the SVM classification towards the majority class (interictal states). The SMOTE algorithm is implemented iteratively. For each minority class sample, its k nearest neighbors are first determined, from which N nearest neighbors are randomly selected. Each sample is subtracted from the selected nearest neighbor and multiplied by a factor in the range (0, 1) to create a new synthetic sample. In this scheme, k is set to 5, and N is adaptively set to the integer ratio between the number of majority and minority class samples. If N is greater than k, then N = k. Finally, the augmented data is fed into a support vector machine for classification into ictal and interictal states. The classifier is set with a second-order polynomial kernel function. Training and validation are performed using five-fold cross-validation.

[0095] This application is able to achieve excellent performance in automatic epileptic seizure detection technology. The key lies in the innovative use of parameterization technology to obtain non-periodic and periodic component features. On the one hand, epileptologists provide prior knowledge about the ictal period and the interictal period. On the other hand, electrophysiologists further reveal the connection between the EEG power spectrum and neural activity. Based on the above two points, this solution uses a hybrid model that combines the Lorentz function and the multi-Gaussian function to parameterize the power spectrum of the epileptic EEG to obtain its non-periodic and periodic components; based on the statistical analysis of the system, four parameterized features are finally selected: the offset and index of the non-periodic component, and the characteristic total power at the frequency corresponding to the two maximum powers in the periodic component; for specific patients, the representative channel is adaptively selected to finally realize an automatic epileptic seizure monitoring system based on a single channel.

[0096] Compared to the existing techniques in Part 1, the greatest advantage of this technique lies in its ability to adapt to the EEG power spectrum of different patients. For example, Swinnen et al. extracted 13 features in the time and frequency domains, all extracted within or between pre-specified fixed frequency bands; Janjarasjitt et al. used eight wavelet-based features to identify the most common frequency subbands in epileptic EEG signals; Lu Yanan et al. extracted 12 features based on the Hilbert-Huang transform and Kraskov entropy, some of which were adaptive and some not; and Romanie calculated envelope-based phase synchronization features in the β-12-30 Hz frequency band, which has low frequency correlation with ictal EEG signals. Because this technique's feature extraction does not require pre-specified frequency subbands and can adaptively select representative channels for a specific patient, it is highly tolerant of individual differences in the power spectrum. Therefore, this technique can achieve consistent performance for each patient, indicating a highly robust monitoring model.

[0097] In addition, this application also proposes a long-term monitoring device for epileptic seizures, such as Figure 5 As shown, the device includes:

[0098] The first module is used to obtain EEG data of all channels;

[0099] The second module is used to extract features using segmented EEG PSD parameterization, where the non-periodic component is characterized by two parameters: offset and exponent, and the periodic component is characterized by the two maximum powers PW1 and PW2 and their corresponding center frequencies CF1 and CF2. The characteristic total powers TPW1 and TPW2 correspond to the original PSD values corresponding to the center frequencies CF1 and CF2, respectively.

[0100] The third module is used to select several key parameterized features that are most relevant to the seizure event;

[0101] A fourth module is configured to select the most representative channel based on the statistically significant P values of the key parameterized features during the ictal and interictal periods;

[0102] The fifth module is used to automatically detect epileptic seizures based on the support vector machine classifier according to the EEG signal features of the most representative channels.

[0103] In addition, this application also proposes a long-term monitoring device for epileptic seizures, such as Figure 6 As shown, the device includes:

[0104] A single-channel EEG data acquisition device;

[0105] at least one processor;

[0106] At least one memory, the memory being used to store at least one program and EEG data collected by the above-mentioned device;

[0107] When at least one of the programs is executed by at least one of the processors, the long-term monitoring method for epileptic seizures as described in the previous embodiment is implemented.

[0108] The contents of the above method embodiments are all applicable to the present device embodiments. The functions specifically implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0109] In addition, the present application also proposes a storage medium, which stores a program executable by a processor. When the program executable by the processor is executed by the processor, the long-term monitoring method for epileptic seizures as described in the previous embodiment is implemented.

[0110] Similarly, the contents of the above method embodiments are applicable to the present storage medium embodiment. The functions specifically implemented by the present storage medium embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0111] In some optional embodiments, the functions / operations mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the functions / operations involved, the two boxes shown in succession may actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiments presented and described in the flow chart of the present application are provided in an exemplary manner for the purpose of providing a more comprehensive understanding of the technology. The disclosed method is not limited to the operations and logic flows presented herein. Optional embodiments are contemplated in which the order of the various operations is changed and the sub-operations described as a part of a larger operation are performed independently.

[0112] In addition, although the present application is described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in separate physical devices or software modules. It is also understood that a detailed discussion of the actual implementation of each module is not necessary for understanding the present application. More specifically, given the properties, functions, and internal relationships of the various functional modules in the devices disclosed herein, the actual implementation of the module will be understood within the routine skills of an engineer. Therefore, a person skilled in the art can implement the present application as set forth in the claims using ordinary techniques without undue experimentation. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present application, which is determined by the full scope of the appended claims and their equivalents.

[0113] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0114] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0115] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0116] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0117] In the above description of this specification, reference to the terms "one embodiment / example," "another embodiment / example," or "certain embodiments / examples" means that the specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples.

[0118] Although the embodiments of the present application have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions, and variations may be made to the embodiments without departing from the principles and intent of the present application, and that the scope of the present application is defined by the claims and their equivalents.

[0119] The above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A long-term monitoring device for epileptic seizures, characterized in that: The device comprises: The first module is used to obtain data from all channels; filter the data from all channels and perform segmented preprocessing of EEG signals; Among them, the data of all channels are filtered and the EEG signal is segmented and pre-processed, including: The EEG data were filtered using a high-pass filter and a low-pass filter with zero phase shift; The preictal and postictal periods were defined as the 10 minutes before and after the ictal period, respectively, and the part excluding the preictal, postictal, and ictal periods was defined as the interictal period; Sliding 5-second time windows were used to segment ictal and interictal EEG signals without overlap, and the DC component of the EEG signals was removed; The cutoff frequency of the high-pass filter is -3dB, 0.5Hz, and the cutoff frequency of the low-pass filter is -3dB, 45Hz; The second module is used to extract features using PSD parameterization, where the non-periodic component is characterized by two parameters: offset and exponent, and the periodic component is characterized by the two maximum powers PW1 and PW2 and their corresponding center frequencies CF1 and CF2. The characteristic total powers TPW1 and TPW2 correspond to the original PSD values corresponding to the center frequencies CF1 and CF2, respectively. The third module is used to select several key parameterized features that are most relevant to the seizure event; Among them, the key parameterized features include four items: offset, index, TPW1 and TPW2; A fourth module is configured to select the most representative channel based on the statistically significant P values of the key parameterized features during the ictal and interictal periods; Among them, several key parameterized features with the highest correlation with the seizure event are selected, and the most representative channels are selected according to the statistical significance P value of the key parameterized features during the seizure period and the interictal period, including: Ranking based on the statistical significance P value of key parameters in the interictal and ictal periods; Select the most representative channels based on the ranking; After selecting the most representative channel according to the statistically significant P values of the key parameterized features during the ictal and interictal periods, the fourth module further performs the following operations: Performing data enhancement on the EEG data of the most representative channel; The fifth module is used to automatically detect epileptic seizures based on the support vector machine classifier according to the four features of the most representative channel EEG; the four features include the offset and index of the non-periodic component, and the characteristic total power TPW1 and TPW2; Among them, the automatic detection of epileptic seizures is performed based on the support vector machine classifier according to the above four features of the most representative channels, including the following data augmentation methods: Determine the 5 nearest neighbors of the sample and randomly select N nearest neighbors from them; Each sample is subtracted from the selected neighboring sample and multiplied by a random factor between 0 and 1 to obtain a new synthetic sample; The amplified data were input into the support vector machine to classify the ictal and interictal states; Here, N is adaptively set to an integer ratio between the number of samples in the majority class and the minority class.

2. The long-term monitoring device for epileptic seizures according to claim 1, characterized in that: The fitting calculation formula of the PSD function is: Where PSD is the power spectral density (PSD) function, L is the non-periodic component, G is the periodic component, b is the offset, is the index, F is the input frequency vector, ranging from 0.5 to 30 Hz, with an interval of 0.5 Hz, k is the "inflection point" parameter, which is used to control the bending of the non-periodic component and is set to 0 in this application, and N is the number of Gaussian function combinations, with a maximum value of 8 in this application. is the peak power, expressed as log10(Power), and is the center frequency and standard deviation of the Gaussian function.

3. A storage medium, characterized in that The storage medium stores a program executable by a processor, and when the program executable by the processor is executed by the processor, the long-term monitoring device for epileptic seizures according to any one of claims 1 to 2 is implemented.

Images