CN111938631B - Convulsion detection method based on neonatal electroencephalogram signal - Google Patents
Convulsion detection method based on neonatal electroencephalogram signal Download PDFInfo
- Publication number
- CN111938631B CN111938631B CN202010788514.3A CN202010788514A CN111938631B CN 111938631 B CN111938631 B CN 111938631B CN 202010788514 A CN202010788514 A CN 202010788514A CN 111938631 B CN111938631 B CN 111938631B
- Authority
- CN
- China
- Prior art keywords
- data
- convulsion
- peak
- data segment
- average
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/40—Detecting, measuring or recording for evaluating the nervous system
- A61B5/4058—Detecting, measuring or recording for evaluating the nervous system for evaluating the central nervous system
- A61B5/4064—Evaluating the brain
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7203—Signal processing specially adapted for physiological signals or for diagnostic purposes for noise prevention, reduction or removal
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/725—Details of waveform analysis using specific filters therefor, e.g. Kalman or adaptive filters
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B5/00—Measuring for diagnostic purposes; Identification of persons
- A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
- A61B5/7235—Details of waveform analysis
- A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
- A61B5/7267—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems involving training the classification device
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B2503/00—Evaluating a particular growth phase or type of persons or animals
- A61B2503/04—Babies, e.g. for SIDS detection
- A61B2503/045—Newborns, e.g. premature baby monitoring
Landscapes
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Physics & Mathematics (AREA)
- Biophysics (AREA)
- Public Health (AREA)
- Veterinary Medicine (AREA)
- Physiology (AREA)
- General Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- Animal Behavior & Ethology (AREA)
- Surgery (AREA)
- Pathology (AREA)
- Biomedical Technology (AREA)
- Heart & Thoracic Surgery (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Psychiatry (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Neurology (AREA)
- Evolutionary Computation (AREA)
- Fuzzy Systems (AREA)
- Mathematical Physics (AREA)
- Psychology (AREA)
- Neurosurgery (AREA)
- Measurement And Recording Of Electrical Phenomena And Electrical Characteristics Of The Living Body (AREA)
Abstract
The invention discloses a convulsion detection method based on neonatal electroencephalogram signals, which comprises a training part and a testing part, wherein the training part is used for filtering data by 0.5-40 Hz, then carrying out down-sampling of 32Hz, then carrying out filtering of 0.5-12.5 Hz, then dividing all data into 24s sections, extracting and standardizing characteristic values, carrying out classification marking on the characteristic values, and training by using an SVM (support vector machine) to obtain a model; and the test part filters the data at 0.5-40 Hz, performs down sampling at 32Hz, performs filtering at 0.5-12.5 Hz, extracts a characteristic value for every 24s of data, eliminates non-convulsion data according to a threshold condition, normalizes the data which cannot be eliminated, predicts by using a model and judges whether the data is convulsion data. The method can detect the convulsion of the neonatal electroencephalogram signal, can assist medical personnel in diagnosing the convulsion of the neonatal, and improves the diagnosis efficiency.
Description
Technical Field
The invention relates to a convulsion detection method based on neonatal electroencephalogram signals, and belongs to an image processing technology.
Background
Convulsion is the abnormal discharge of brain cells caused by the temporary disorder of cerebral cortex function, which is manifested by the sudden occurrence of involuntary myotonia and clonic convulsion in the systemic or local skeletal muscle group and the resulting joint movement, mostly systemic and symmetric, often accompanied by disturbance of consciousness. Convulsions can be caused by abnormal discharges in the central nervous system only, but such abnormal discharges can be caused by many primary intracranial lesions (meningitis, cerebrovascular accidents, encephalitis, intracranial haemorrhages, tumors), or secondary to systemic or metabolic (e.g. ischemia, hypoxia, hypoglycemia, hypocalcemia, hyponatremia) diseases. Complications such as asphyxia and cerebral anoxia easily occur in neonatal convulsion, and care, prevention and health care of neonates should be closely paid attention.
Therefore, the detection significance for the neonatal convulsion is important, the detection equipment for the neonatal convulsion is not common at present, and the neonatal electroencephalogram is monitored to detect whether the neonatal convulsion occurs or not without precedent.
Reference documents
【1】Temko,A.,et al.,EEG-based neonatal seizure detection with Support Vector Machines.Clinical Neurophysiology,2011.122(3):p.464-473.
【2】Gotman,J.,et al.,Automatic seizure detection in the newborn:methods and initial evaluation.Electroencephalography and Clinical Neurophysiology,1997.103(3):p.356-362.
【3】Temko,A.,et al.,Clinical implementation of a neonatal seizure detection algorithm.Decision Support Systems,2015.70:p.86-96.
Disclosure of Invention
The invention aims to: in order to overcome the defects of the prior art, the invention provides a convulsion detection method based on a neonatal electroencephalogram signal, which can detect whether a neonate has convulsion and prevent the neonate from suffering from the convulsion.
The technical scheme is as follows: in order to achieve the purpose, the invention adopts the technical scheme that:
a convulsion detection method based on neonatal brain electrical signals comprises a training phase and a detection phase, and specifically comprises the following steps:
a training stage:
s11, collecting electroencephalogram signals of P (P is more than 50) convulsion newborns as samples, and performing band-pass filtering for removing error data; the original electroencephalogram signals contain a large amount of noise, and can not be directly interpreted manually, wherein a band-pass filter for eliminating error data is firstly carried out, and then manual interpretation can be carried out; preferably, 0.5-40 Hz band-pass filtering is adopted, of course, the given cut-off frequency is not fixed, and 0.5-70 Hz band-pass filtering can also be adopted, or the quality of the electroencephalogram signal is determined;
s12, performing down-sampling on all samples obtained in the step S11, preferably adopting 32Hz down-sampling; performing band-pass filtering for screening the calculated data, wherein the band-pass filtering is to remove unnecessary signals, and preferably 0.5-12.5 Hz (reference 1);
s13, dividing each sample obtained in the step S12 into data segments of one ks segment, and discarding the data segments which are less than ks;
s14, identifying all data sections, dividing the data sections into three data sets according to the existence of convulsion, wherein the three data sets are a convulsion data set, a non-convulsion data set and other data sets respectively, all convulsion data form the convulsion data set, all non-convulsion data form the non-convulsion data set, and ks data sections except the convulsion data and the non-convulsion data form other data sets; the convulsion data is a ks data section with a convulsion signal appearing every second in the ks data section, and the non-convulsion data is a ks data section without a convulsion signal appearing every second in the ks data section;
s15, marking the convulsion data as a and the non-convulsion data as b;
s16, extracting characteristic values of all convulsion data and non-convulsion data;
s17, standardizing all the characteristic values to obtain the average value and the variance of normal distribution;
s18, randomly selecting Q (Q is more than or equal to 1000 and is not more than 30 percent of all data) data from the convulsion data and the non-convulsion data respectively for the normalized characteristic values, and selecting all or part of the characteristic values extracted in the step S16 to carry out classification training to obtain a classification Model;
and (3) a testing stage:
s21, performing band-pass filtering processing for eliminating error data on the collected neonatal electroencephalogram signals;
s22, down-sampling the data obtained in the step S21, calculating band-pass filtering for data screening, and dividing the obtained data into a detection part and a non-detection part;
s23, if the length of the undetected part is greater than or equal to ks, extracting a front ks data segment of the undetected part;
s24, extracting characteristic values of the ks data segments extracted in the step S23;
s25, performing preliminary judgment on the characteristic value extracted in the step S24, and confirming whether the ks data segment extracted in the step S23 is non-convulsion data or not;
s26, if the feature value extracted in step S24 cannot be determined as non-convulsive data in step S25, normalizing the feature value, and predicting using the classification Model obtained in step S18: if the prediction is a, the ks data segment extracted in the step S23 is regarded as convulsion data; if b is predicted, the ks data segment extracted in the step S23 is regarded as non-convulsion data;
s27, marking the ks data segments extracted in the step S23 as detection parts and removing the ks data segments from the non-detection parts;
s28, looping the steps S23 to S27 until the length of all the detected data or the length of the undetected part is less than ks.
Preferably, in step S18, a classification Model is obtained by performing classification training using a support vector machine, a decision tree, a neural network, and the like, wherein the classification accuracy of the support vector machine is higher; if a support vector machine is adopted for classification training, the model used in the training is an SVM classification model (support vector machine classification model), the kernel function is an RBF kernel function (radial basis kernel function), the penalty coefficient of the kernel function is 10, and the gamma function gamma of the kernel function is the reciprocal of the total number of the feature vectors.
Preferably, in step S16, for any ks data segment, the feature value is extracted as follows:
(a) Frequency domain main peak average frequency: firstly, dividing a ks data segment into m k/ms data segments; then calculating the power density spectrum of each k/ms data segment, and taking the peak with the maximum amplitude in each k/ms data segment as the main peak of the k/ms data segment to obtain the corresponding main peak frequency; finally, averaging the m main peak frequencies, namely the frequency domain main peak average frequency of the ks data segment; wherein m is a positive integer;
(b) Mean frequency standard deviation of main peak in frequency domain: based on the step (a), calculating the standard deviation of the m main peak frequencies, namely the frequency domain main peak average frequency standard deviation of the ks data segment;
(c) Frequency domain main peak average wave width: based on the step (a), averaging the half wave widths corresponding to the m main peaks, namely, the average wave width of the main peak of the frequency domain of the ks data segment;
(d) Frequency domain main peak power ratio average: based on the step (a), calculating power of a half-wave-width area corresponding to each main peak, taking the ratio of the power to the total power of the k/ms data segment as the main peak power ratio of the k/ms data segment, and averaging m main peak power ratios to obtain the average value of the frequency-domain main peak power ratio of the ks data segment;
(e) Time domain average peak height: dividing a ks data segment into n sections of k/ns data segments; then taking the highest peak height of each k/ns data segment as a peak extreme value; finally, averaging the n peak extreme values to obtain the time domain average peak height of the ks data segment; wherein n is a positive integer greater than m;
(f) Height stability of time domain peak: based on step (e), taking the maximum/minimum value of the n peak extreme values as the time domain peak height stability of the ks data segment;
(g) Time domain average peak width: based on the step (e), averaging the half wave widths corresponding to the n peak extremums to obtain the time domain average peak width of the ks data segment;
(h) Stability of time domain peak width: and (e) based on the step (e), taking the maximum value/minimum value in the half wave width corresponding to the n peak extremum values as the time domain peak width stability of the ks data segment.
Preferably, in step S18, after two feature values of the time domain peak height stability and the time domain peak width stability are removed, the remaining feature values are classified and trained to obtain a classification Model; during testing, the two items are eliminated, and the accuracy of the classification Model cannot be greatly influenced, so that the two items are eliminated during training, and the performance problem is mainly considered.
Preferably, in the step S25, the determination is performed according to the following conditions (the threshold value in the following determination conditions is variable, and is determined after the test), and if one or more of the following conditions is satisfied, the ks data segment extracted in the step S23 is regarded as the non-convulsion data:
(1) frequency domain main peak average frequency f m ：f m <0.5Hz or f m >5Hz；
(2) Mean frequency standard deviation sigma of frequency domain main peak m ：σ m >7Hz；
(3) Frequency domain main peak average bandwidth w f ：w f >2Hz；
(4) Frequency domain main peak power ratio average r: r <0.2 or r >0.8;
(5) time domain average peak height h t ：h t <20 μ V or h t >250μV
(6) Time domain peak height stability s 1 ：s 1 <1 or s 1 >6；
(7) Time domain average peak width w t ：w t <0.18s or w t >1s；
(8) Time domain peak width stability s 2 ：s 2 <1 or s 2 >6。
Preferably, k is 24, 10 or 8, and after multiple tests, the detection result obtained by adopting the value of k =24 can better meet the application requirement; m =3, which is a more preferred value for k = 24; n =12, which is a preferred value for k =24, because the frequency of the convulsive signal is generally 0.5 to 5hz, and the time domain period of n =12 is 2s, and if there is convulsion, there is always a convulsive signal in the time domain period.
Has the advantages that: the convulsion detection method based on the neonatal electroencephalogram signals can be used for detecting convulsions of the neonatal electroencephalogram signals, can assist medical staff in diagnosing the convulsions of the neonates, and improves diagnosis efficiency.
Drawings
Both (a) and (b) in fig. 1 are brain electrical signal samples with convulsion;
both (a) and (b) in fig. 2 are normal brain electrical signal samples;
FIG. 3 is a schematic diagram of a training process;
FIG. 4 is a schematic diagram of a test flow;
Detailed Description
The invention is described in detail below with reference to the figures and the embodiments.
In the description of the present invention, it is to be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", and the like, indicate orientations or positional relationships based on those shown in the drawings, and are used only for convenience in describing the present invention and for simplicity in description, and do not indicate or imply that the referenced devices or elements must have a particular orientation, be constructed and operated in a particular orientation, and thus, are not to be construed as limiting the present invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance.
A convulsion detection method based on neonatal brain electrical signals comprises a training stage and a detection stage, and specifically comprises the following steps:
a training stage:
s11, collecting electroencephalogram signals of P convulsion newborns as samples, and performing band-pass filtering at 0.5-40 Hz;
s12, carrying out down-sampling of 32Hz on all samples obtained in the step S11, and carrying out band-pass filtering of 0.5-12.5 Hz;
s13, dividing each sample obtained in the step S12 into 24S data segments, and discarding the data segments less than 24S;
s14, identifying all data sections, dividing the data sections into three data sets according to the existence of convulsion, wherein the three data sets are a convulsion data set, a non-convulsion data set and other data sets respectively, all convulsion data form the convulsion data set, all non-convulsion data form the non-convulsion data set, and 24S data sections except the convulsion data and the non-convulsion data form other data sets; the convulsion data is a 24s data segment with a convulsion signal appearing every second in the 24s data segment, and the non-convulsion data is a 24s data segment without the convulsion signal appearing every second in the 24s data segment;
s15, marking the convulsion data as 1, and marking the non-convulsion data as-1;
s16, extracting characteristic values of all convulsion data and non-convulsion data; for any 24s data segment, extracting characteristic values according to the following modes:
(a) Frequency domain main peak average frequency: firstly, dividing a 24s data section into three 8s data sections; then calculating the power density spectrum of each section of 8s data segment, and taking the peak with the maximum amplitude in each section of 8s data segment as the main peak of the 8s data segment to obtain the corresponding main peak frequency; finally, averaging the three main peak frequencies, namely the frequency domain main peak average frequency of the 24s data segment;
(b) Mean frequency standard deviation of main peak in frequency domain: based on the step (a), calculating the standard deviation of the three main peak frequencies, namely the standard deviation of the frequency domain main peak average frequency of the 24s data segment;
(c) Average wave width of main peak in frequency domain: based on the step (a), averaging the half wave widths corresponding to the three main peaks, namely the frequency domain main peak average wave width of the 24s data segment;
(d) Frequency domain main peak power ratio average: based on the step (a), calculating power of a half-wave width region corresponding to each main peak, taking the ratio of the power to the total power of the affiliated 8s data segment as the main peak power ratio of the 8s data segment, and averaging the three main peak power ratios to obtain the frequency domain main peak power ratio average value of the 24s data segment;
(e) Time domain average peak height: firstly, dividing a 24s data section into twelve sections of 2s data sections; then taking the highest peak height of each 2s data segment as a peak extreme value; finally, averaging twelve peak extreme values to obtain the time domain average peak height of the 24s data segment;
(f) Height stability of time domain peak: based on step (e), taking the maximum/minimum of the twelve peak extrema as the temporal peak height stability of the 24s data segment;
(g) Time domain average peak width: based on the step (e), averaging the half wave widths corresponding to the twelve peak extremum values to obtain the time domain average peak width of the 24s data segment;
(h) Stability of time domain peak width: based on the step (e), taking the maximum value/minimum value in the half wave widths corresponding to the twelve peak extrema as the stability of the time domain peak width of the 24s data segment;
s17, standardizing all the characteristic values to obtain the average value and the variance of normal distribution;
s18, respectively and randomly selecting 1000 pieces of data from the convulsive data and the non-convulsive data as training data for the normalized characteristic values, and after removing two characteristic values of the height stability and the width stability of the time domain wave peak, carrying out classification training by using a support vector machine to obtain a classification Model; the model used in training is an SVM classification model (support vector machine classification model), the kernel function is an RBF kernel function (radial basis kernel function), the penalty coefficient of the kernel function is 10, and the gamma function of the kernel function is 1/6 (gamma = 1/6);
and (3) a testing stage:
s21, filtering the collected neonatal electroencephalogram signals by using a 0.5-40 Hz band-pass filter;
s22, carrying out down-sampling of 32Hz on the data obtained in the step S21, carrying out band-pass filtering of 0.5-12.5 Hz, and dividing the obtained data into a detection part and an undetected part;
s23, if the length of the undetected part is more than or equal to 24S, extracting the first 24S data segment of the undetected part;
s24, extracting characteristic values of the 24S data segments extracted in the step S23;
and S25, judging the characteristic value extracted in the step S24 according to the following conditions, and if more than one of the following conditions is met, considering the 24S data segment extracted in the step S23 as non-convulsion data:
(1) frequency domain dominant peak average frequency f m ：f m <0.5Hz or f m >5Hz；
(2) Mean frequency standard deviation sigma of frequency domain main peak m ：σ m >7Hz；
(3) Frequency domain main peak average bandwidth w f ：w f >2Hz；
(4) Frequency domain main peak power ratio average r: r <0.2 or r >0.8;
(5) time domain average peak height h t ：h t <20 μ V or h t >250μV
(6) Time domain peak height stability s 1 ：s 1 <1 or s 1 >6；
(7) Time domain average peak width w t ：w t <0.18s or w t >1s；
(8) Time domain peak width stability s 2 ：s 2 <1 or s 2 >6；
S26, if the feature value extracted in step S24 does not satisfy any of the conditions given in step S25, normalizing the feature value extracted in step S24, and predicting using the classification Model obtained in step S18: if the prediction is 1, the 24S data segment extracted in the step S23 is regarded as convulsion data; if the prediction is-1, the 24S data segment extracted in the step S23 is regarded as non-convulsion data;
s27, marking the 24S data segment extracted in the step S23 as a detection part, and removing the data segment from the non-detection part;
s28, looping the steps S23 to S27 until the length of the detected or undetected part of all data is less than 24S.
In the description of the present invention, it should be noted that, unless otherwise explicitly specified or limited, the terms "mounted," "connected," and "connected" are to be construed broadly and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
The foregoing shows and describes the general principles, principal features and advantages of the invention. It should be understood by those skilled in the art that the above embodiments do not limit the present invention in any way, and all technical solutions obtained by using equivalent alternatives or equivalent variations fall within the scope of the present invention.
Claims (4)
1. A convulsion detection method based on neonatal electroencephalogram signals is characterized in that: the method comprises a training stage and a detection stage, and specifically comprises the following steps:
a training stage:
s11, collectingPTaking the electroencephalogram signals of the convulsion neonates as samples, and performing band-pass filtering for removing error data;
s12, performing down-sampling on all samples obtained in the step S11, and performing band-pass filtering for data screening;
s13, dividing each sample obtained in the step S12 intokData segment of one second, deficiencykData segments of seconds are discarded;
s14, identifying all data segments, dividing the data segments into three data sets according to the existence of convulsion, namely a convulsion data set, a non-convulsion data set and other data sets, wherein all convulsion data form the convulsion data set, all non-convulsion data form the non-convulsion data set, and all convulsion data and non-convulsion data are not includedkThe second data segment forms other data sets; the convulsion data arekThe convulsive signal occurring every second of the second data segmentkSecond data segment, non-convulsive data ofkWith no convulsive signal occurring every second of the second data segmentkA second data segment;
s15, marking convulsion data asaNon-convulsive data markersb；
S16, extracting characteristic values of all convulsion data and non-convulsion data; for any one ofkAnd (3) extracting a characteristic value from the second data segment according to the following steps:
（a) Frequency domain main peak average frequency: firstly, the method is tokSecond data segment separationmSegment ofk/mA second data segment; recalculate each segmentk/mPower density spectrum of second data segment, each segmentk/mNumber of secondsTaking the peak with the maximum amplitude in the segment as thek/mObtaining the corresponding main peak frequency by the main peak of the second data segment; last pairmThe average value of the main peak frequency is the average valuekThe frequency domain main peak average frequency of the second data segment; wherein the content of the first and second substances,mis a positive integer;
（b) Mean frequency standard deviation of main peak in frequency domain: based on the steps of (a) CalculatingmThe standard deviation of the frequency of each main peak is the frequency of the main peakkThe frequency domain main peak average frequency standard deviation of the second data segment;
（c) Average wave width of main peak in frequency domain: based on the steps of (a) To, formThe half wave width corresponding to each main peak is averaged to obtain the average valuekThe frequency domain main peak average wave width of the second data segment;
（d) Frequency domain main peak power ratio average: based on the steps of (a) Calculating power of half-wave width region corresponding to each main peak, and comparing the power with the corresponding main peakk/mThe ratio of the total power of the second data segment is used as the ratiok/mRatio of main peak power to second data segmentmThe average value of the ratio of the main peak power is the average valuekThe frequency domain main peak power ratio average value of the second data segment;
（e) Time domain average peak height: firstly, the method is carried outkSeparation of second data segment intonSegment ofk/nA second data segment; then each segment is divided intok/nThe highest peak height of the second data segment is used as a peak extreme value; last pair ofnThe peak extreme value is averaged to obtain the average valuekTime domain average peak height of the second data segment; wherein, the first and the second end of the pipe are connected with each other,nis greater thanmA positive integer of (d);
（f) Height stability of time domain peak: based on the steps of (e) Will benThe maximum/minimum value of the peak extreme values is used as the maximum/minimum valuekThe time domain peak height stability of the second data segment;
（g) Time domain average peak width: based on the steps of (e) To is aligned withnAveraging the half wave widths corresponding to the extreme values of the wave crests to obtain the average valuekTime domain average peak width of the second data segment;
（h) Stability of time domain peak width: based on the steps of (e) Will benThe maximum value/minimum value in the half-wave width corresponding to the peak extreme value is used as the maximum value/minimum valuekStability of time domain peak width of the second data segment;
s17, standardizing all the characteristic values, and calculating the average value and the variance of normal distribution to obtain the standardized characteristic values;
s18, respectively randomly selecting convulsion data and non-convulsion dataQTaking the bar data as training data, and selecting all or part of the normalized characteristic values to perform classification training to obtain a classification Model; carrying out classification training by using a support vector machine to obtain a classification Model; the model used in training is an SVM classification model, the kernel function is an RBF kernel function, and the penalty coefficient of the kernel function is 10;
and (3) a testing stage:
s21, performing band-pass filtering processing for eliminating error data on the collected neonatal electroencephalogram signals;
s22, down-sampling the data obtained in the step S21, performing band-pass filtering for data screening, and dividing the obtained data into a detection part and an undetected part;
s23, if the length of the undetected part is larger than or equal tokSecond, before extracting the undetected portionkA second data segment;
s24, extracting from the step S23kExtracting characteristic values of the second data segment;
s25, performing preliminary judgment on the characteristic value extracted in the step S24, and confirming the characteristic value extracted in the step S23kWhether the second data segment is non-convulsive data or not; the judgment is made under the condition that one or more of the following conditions are satisfied, and it is considered that the result is extracted in step S23kThe second data segment is non-convulsive data:
(1) frequency domain main peak average frequencyf m ：f m < 0.5Hz orf m ＞5Hz；
(2) Frequency domain main peak average frequency standard deviationσ m ：σ m ＞7Hz；
(3) Frequency domain main peak average bandwidthw f ：w f ＞2Hz；
(4) Frequency domain main peak power ratio average valuer：r<0.2 orr＞0.8；
(5) Time domain average peak heighth t ：h t ＜20μV orh t ＞250μV；
(6) Time domain peak height stabilitys 1 ：s 1 < 1 ors 1 ＞6；
(7) Time domain average peak widthw t ：w t < 0.18 seconds orw t More than 1 second;
(8) time domain peak width stabilitys 2 ：s 2 < 1 ors 2 ＞6；
S26, if the feature value extracted in step S24 cannot be determined as non-convulsive data in step S25, normalizing the feature value, and predicting using the classification Model obtained in step S18: if it is predicted asaThen, the step S23 is considered as the extracted stepkThe second data segment is convulsion data; if it is predicted asbThen, the step S23 is considered as the extracted stepkThe second data segment is non-convulsion data;
s27, extracting step S23kMarking the second data segment as a detection part and removing the second data segment from an undetected part;
s28, the steps S23 to S27 are repeated until the length of all the detected data or the length of the undetected part is less thankAnd seconds.
2. A method of detecting convulsions based on a neonatal brain electrical signal according to claim 1, wherein: in step S18, after two feature values of the time domain peak height stability and the time domain peak width stability are removed, the remaining feature values are subjected to classification training to obtain a classification Model.
3. A method of detecting convulsions based on a neonatal brain electrical signal according to claim 1, wherein:k=24，m=3，n=12。
4. a method of detecting convulsions based on a neonatal brain electrical signal according to claim 1, wherein: band-pass filtering for eliminating error data, wherein the filtering range is 0.5 to 40Hz; the band-pass filtering of the calculation data screening is carried out, and the filtering range is 0.5-12.5 Hz; the down-sampling has a frequency of 32Hz.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010788514.3A CN111938631B (en) | 2020-08-07 | 2020-08-07 | Convulsion detection method based on neonatal electroencephalogram signal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010788514.3A CN111938631B (en) | 2020-08-07 | 2020-08-07 | Convulsion detection method based on neonatal electroencephalogram signal |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111938631A CN111938631A (en) | 2020-11-17 |
| CN111938631B true CN111938631B (en) | 2023-01-31 |
Family
ID=73331719
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010788514.3A Active CN111938631B (en) | 2020-08-07 | 2020-08-07 | Convulsion detection method based on neonatal electroencephalogram signal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111938631B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114711790B (en) * | 2022-04-06 | 2022-11-29 | 复旦大学附属儿科医院 | Newborn electroconvulsive type determination method, newborn electroconvulsive type determination device, newborn electroconvulsive type determination equipment and storage medium |
| CN115568867B (en) * | 2022-11-15 | 2023-11-10 | 南京左右脑医疗科技集团有限公司 | Treatment efficacy evaluation method, device and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008058343A1 (en) * | 2006-11-15 | 2008-05-22 | The University Of Queensland | A method for detecting eeg seizures in a newborn or a young child |
| CN104173046A (en) * | 2014-09-01 | 2014-12-03 | 清华大学 | Method for extracting color marked amplitude-integrated electroencephalogram |
| CN109300544A (en) * | 2018-12-10 | 2019-02-01 | 南京伟思医疗科技股份有限公司 | A kind of seven step analysis method of newborn baby function standardizing |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB0906029D0 (en) * | 2009-04-07 | 2009-05-20 | Nat Univ Ireland Cork | A method of analysing an electroencephalogram (EEG) signal |
-
2020
- 2020-08-07 CN CN202010788514.3A patent/CN111938631B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008058343A1 (en) * | 2006-11-15 | 2008-05-22 | The University Of Queensland | A method for detecting eeg seizures in a newborn or a young child |
| CN104173046A (en) * | 2014-09-01 | 2014-12-03 | 清华大学 | Method for extracting color marked amplitude-integrated electroencephalogram |
| CN109300544A (en) * | 2018-12-10 | 2019-02-01 | 南京伟思医疗科技股份有限公司 | A kind of seven step analysis method of newborn baby function standardizing |
Non-Patent Citations (4)
| Title |
|---|
| A comparison of quantitative EEG features for neonatal seizure detection;B.R. Greene 等;《Clinical Neurophysiology》;20080401;全文 * |
| A Multi-stage System for the Automated Detection of Epileptic Seizures in Neonatal EEG;Joyeeta Mitra 等;《J Clin Neurophysiol》;20090831;全文 * |
| An SVM-Based System and Its Performance for Detection of Seizures in Neonates;Andriy Temko;《31st Annual International Conference of the IEEE EMBS》;20090906;全文 * |
| Automatic seizure detection in the newborn: methods and initial evaluation;J. Gotman;《Electroencephalography and clinical Neurophysiology》;19970217;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111938631A (en) | 2020-11-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111938631B (en) | Convulsion detection method based on neonatal electroencephalogram signal | |
| George et al. | Classification of epileptic EEG signals using PSO based artificial neural network and tunable-Q wavelet transform | |
| US8364255B2 (en) | Method and device for removing EEG artifacts | |
| US20200022603A1 (en) | Semiology of seizures including muscle signals collected from electroencephalography electrodes | |
| US8700141B2 (en) | Method and apparatus for automatic evoked potentials assessment | |
| US20080167569A1 (en) | Processing of Physiological Signal Data in Patient Monitoring | |
| US20040133115A1 (en) | Method and apparatus for identifying heart rate feature events | |
| US20070197930A1 (en) | Detection of epileptiform activity | |
| CN102973266B (en) | For method, equipment and computer program that automatic spasm is monitored | |
| Petersen et al. | Generic single-channel detection of absence seizures | |
| CN111093471B (en) | Method for detecting pathological brain activity from scalp electroencephalograms | |
| DK177536B1 (en) | Method for detecting seizures | |
| WO2005058156A1 (en) | A system and a method for analysing ecg curvature for long qt syndrome and drug influence | |
| Chaibi et al. | A comparaison of methods for detection of high frequency oscillations (HFOs) in human intacerberal EEG recordings | |
| Reed et al. | Automatic detection of periods of slow wave sleep based on intracranial depth electrode recordings | |
| Abbasi et al. | Applications of advanced signal processing and machine learning in the neonatal hypoxic-ischemic electroencephalography | |
| WO2014136704A1 (en) | Brain function activity evaluation device and evaluation system using same | |
| US4611284A (en) | Method for decomposing an electromyogram into individual motor unit action potentials | |
| Thangavel et al. | Improving automated diagnosis of epilepsy from EEGs beyond IEDs | |
| Ieracitano et al. | Wavelet coherence-based clustering of EEG signals to estimate the brain connectivity in absence epileptic patients | |
| CN114532993A (en) | Automatic detection method for electroencephalogram high-frequency oscillation signals of epileptic | |
| Zhao et al. | A knowledge-based approach for automatic quantification of epileptiform activity in children with electrical status epilepticus during sleep | |
| Fani et al. | EEG-based automatic epilepsy diagnosis using the instantaneous frequency with sub-band energies | |
| KR20200142441A (en) | Measurement apparatus of altered cortical region using functional network and method thereof | |
| Yadav et al. | A novel morphology-based classifier for automatic detection of epileptic seizures |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |