CN106264519A - A kind of epileptic electroencephalogram (eeg) detection device based on Stockwell conversion - Google Patents

Abstract

The invention discloses a kind of epileptic electroencephalogram (eeg) detection device based on Stockwell conversion, including eeg amplifier, data collecting card and computer, in described computer in be provided with EEG signals pretreatment module, EEG signals segmentation module, Stockwell transform characteristics extraction module, sort module and post-processing module；The present invention uses Stockwell conversion that the epileptic EEG Signal of time-varying, non-stationary is carried out feature extraction, by the epilepsy in COMPUTER DETECTION EEG signals；This device detection epileptic EEG Signal can reach preferable Detection results, is suitable for online brain electricity classification and Detection.

Description

A kind of epileptic electroencephalogram (eeg) detection device based on Stockwell conversion

Technical field

The present invention relates to a kind of automatic seizure brain electro-detection technical equipment field, particularly to a kind of epileptic electroencephalogram (eeg) detection device based on Stockwell conversion.

Background technology

Epilepsy is a kind of common nervous system disease, is characterized with the supersynchronous electric discharge of cerebral neuron, and has paroxysmal, emergency and the characteristic of interim disordered brain function.Result shows according to incompletely statistics, and the prevalence of epilepsy is 0.5% 2%, and owing to epilepsy is a kind of chronic disease, the epilepsy repeatedly happened suddenly physiologically and is the most all bringing the biggest injury to patient, even threat to life time serious.Diagnosis epilepsy most efficient method is EEG (electrocardiogram) examination, electroencephalogram (Electtroencephalogram, EEG) be from cranium scalp or intracranial recording electrode to the spontaneity of brain neuron group, the electrical activity of rhythmicity.At present, epileptic electroencephalogram (eeg) detection mainly medical personnel relies on experience that electroencephalogram is estimated it and no completes containing characteristic waves such as epileptiform discharge, and its workload is big, easily produces erroneous judgement because of tired.Therefore, in epileptic electroencephalogram (eeg) detects, automatic checkout system can be greatly improved the detection efficiency to EEG.

In recent years, although automatic seizure outbreak time-frequency detection method achieves certain achievement, but still many is had to need improvements.nullSuch as: epilepsy based on DWV (differential windowed variance) detects automatically，Detection sensitivity is 89.3%，The accuracy requirement that cannot meet epileptic electroencephalogram (eeg) detection (sees Majumdar KK,Vardhan P.Automatic seizure detection in ECoG by differential operator and windowed variance.IEEE transactions on neural systems and rehabilitation engineering,19(4):356-365,2011).CN1255320 (CN99124210.6) discloses a kind of method and device automatically determining epileptic discharge threshould recognized by artificial nerve network, the method is it needs to be determined that multiple network parameter and great amount of samples are repeated several times training, training speed is slow, and clinical practice has limitation.

Summary of the invention

In order to overcome defect and the deficiency of prior art, the present invention proposes a kind of epileptic electroencephalogram (eeg) detection device based on Stockwell conversion, use Stockwell conversion that the epileptic EEG Signal of time-varying, non-stationary is carried out feature extraction, by the epilepsy in COMPUTER DETECTION EEG signals, and obtain good testing result.Using this invention detection epileptic EEG Signal time saving and energy saving, accuracy of detection is high, error is little, is suitable for online epilepsy classification and Detection.

Technical scheme is as follows: a kind of epileptic electroencephalogram (eeg) detection device based on Stockwell conversion, including eeg amplifier, data collecting card and computer, in described computer in be provided with EEG signals pretreatment module, EEG signals segmentation module, Stockwell transform characteristics extraction module, sort module and post-processing module.

EEG signals is amplified by described eeg amplifier, is then gathered EEG signals by data collecting card and sends in computer；By computer, EEG signals is processed, pass sequentially through EEG signals pretreatment module, EEG signals segmentation module, Stockwell transform characteristics extraction module, sort module and post-processing module, EEG signals is carried out pretreatment, segmentation, feature extraction, training classification and post-processing operation.

Described eeg amplifier uses Neurofile NT eeg amplifier, and the EEG signals forming EEG signals is amplified；

Described data collecting card uses 16 A/D to change data acquisition board, and frequency acquisition is 256Hz.

Described EEG signals pretreatment module includes using band filter to be filtered EEG signals processing, and frequency range is 0.5-30Hz, to filter the artefact in EEG signal and noise；

Described EEG signals segmentation module, carries out segmentation to the EEG signals after Filtering Processing, and the length of every section of EEG signals is equal, is 1024 points, time a length of 4s.

Described Stockwell transform characteristics extraction module, first carries out Stockwell conversion and obtains S EEG signals x (t) after segmentation_x(τ,f):

$S_x(\mathrm{\τ},f)=\frac{\left|f\right|}{\sqrt{2\mathrm{\π}}}\underset{-\mathrm{\∞}}{\overset{+\mathrm{\∞}}{\∫}}{e}^{-\frac{-{(\mathrm{\τ}-t)}^2{f}^2}{2}}{e}^{-2\mathrm{\π}ift}x\left(t\right)dt$

Wherein, τ is the time, and f is frequency.Then calculating EEG signals power spectral density feature w after Stockwell converts:

W=E{ | S_x(τ,f)|²}

Here, E{ } it is by mathematic expectaion operation.

Described sort module uses Gradient Boosting algorithm, by being trained obtaining grader F to the power spectral density feature of training eeg data, for power spectral density characteristic vector w of one section K lane testing eeg data, calculate this section of brain electricity and belong to the grader output probability p of epilepsy:

$p=\frac{e^{F\left(w\right)}}{e^{F\left(w\right)}+e^{-F\left(w\right)}}.$

Described post-processing module includes Kalman filtering and threshold decision, compares with threshold value and labelling, threshold value=0 after classification output probability p is carried out Kalman filtering, if greater than threshold value, then it is labeled as 1 (epileptic electroencephalogram (eeg)), less than threshold value, is then labeled as 0 (normal brain activity electricity).

By the description of technique scheme, technical solution of the present invention provides the benefit that: Stockwell conversion has higher time frequency resolution compared to other Time-Frequency Analysis Method such as wavelet transformation, Short Time Fourier Transform etc. at signal low frequency, therefore the feature of extraction is more accurate after Stockwell converts, additionally, Stockwell transform operation is simple and quick and is easily achieved；Epilepsy detection device based on Stockwell conversion can meet the demand of epileptic EEG Signal on-line checking, has good real-time and feasibility.

Accompanying drawing explanation

Fig. 1 is the hardware connection diagram of the present invention, wherein: 1. eeg amplifier, 2. data collecting card, 3. computer；

Fig. 2 is one section of EEG signals (a) gathering of the present invention and Stockwell transformation results (b) thereof；

Fig. 3 is EEG signals testing result, wherein Fig. 3 a is the output of grader, Fig. 3 b is the output result after Kalman filtering of grader, and Fig. 3 c is the result (1 represents epileptic electroencephalogram (eeg), and 0 represents normal brain activity electricity) after Kalman filtering after threshold decision.

Detailed description of the invention

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is described, it is clear that described embodiment is only a part of embodiment of the present invention rather than whole embodiments.Based on the embodiment in the present invention, the every other embodiment that those of ordinary skill in the art are obtained under not making creative work premise, broadly fall into the scope of protection of the invention.

According to Fig. 1, Fig. 2 and Fig. 3, the invention provides a kind of epileptic electroencephalogram (eeg) detection device based on Stockwell conversion, including: eeg amplifier 1, data collecting card 2, computer 3.

Specific embodiment: a kind of epileptic electroencephalogram (eeg) detection device based on Stockwell conversion, including eeg amplifier, data collecting card and computer, in described computer in be provided with EEG signals pretreatment module, EEG signals segmentation module, Stockwell transform characteristics extraction module, sort module and post-processing module.

EEG signals is amplified by described eeg amplifier, is then gathered EEG signal by data collecting card and sends in computer；By computer, EEG signals is processed, pass sequentially through EEG signals pretreatment module, EEG signals segmentation module, Stockwell transform characteristics extraction module, sort module and post-processing module, EEG signals is carried out pretreatment, segmentation, feature extraction, training classification and post-processing operation.

Described eeg amplifier uses Neurofile NT eeg amplifier, and the EEG signals forming EEG signals is amplified；

Described data collecting card uses 16 A/D to change data acquisition board, and frequency acquisition is 256Hz.

Described EEG signals pretreatment module includes using band filter to be filtered EEG signals processing, and frequency range is 0.5-30Hz, to filter the artefact in EEG signal and noise；

Described EEG signals segmentation module, carries out segmentation to the EEG signals after Filtering Processing, and the length of every section of EEG signals is equal, is 1024 points, time a length of 4s.

Described Stockwell transform characteristics extraction module, first carries out Stockwell conversion and obtains S EEG signals x (t) after segmentation_x(τ,f):

$S_x(\mathrm{\τ},f)=\frac{\left|f\right|}{\sqrt{2\mathrm{\π}}}\underset{-\mathrm{\∞}}{\overset{+\mathrm{\∞}}{\∫}}{e}^{-\frac{-{(\mathrm{\τ}-t)}^2{f}^2}{2}}{e}^{-2\mathrm{\π}ift}x\left(t\right)dt$

Wherein, τ is the time, and f is frequency.Then calculating EEG signals power spectral density feature w after Stockwell converts:

W=E{ | S_x(τ,f)|²}

Here, E{ } it is by mathematic expectaion operation.

Described sort module uses Gradient Boosting algorithm, by being trained obtaining grader F to the power spectral density feature of training eeg data, for power spectral density characteristic vector w of one section K lane testing eeg data, calculate this section of brain electricity and belong to the grader output probability p of epilepsy:

$p=\frac{e^{F\left(w\right)}}{e^{F\left(w\right)}+e^{-F\left(w\right)}}.$

Described sort module uses Gradient Boosting algorithm, and by being trained obtaining grader to the power spectral density feature of training eeg data, step is as follows:

1) W be training grader used by segmented after eeg data, W={w_i∈R^K, i=1,2 ..., N}, wherein w_iBeing the characteristic vector of i-th section of EEG signals each channel power spectrum density composition, K is the port number of brain electricity, and N is data hop count, and every segment length is LEN=1024 point；Y is correspondence markings amount, Y={y_i∈-1,1}, i=1,2 ..., N}, labelled amount is-1 expression normal brain activity electricity, and labelled amount is 1 expression epileptic electroencephalogram (eeg)；F_mThe grader set up after representing m step iteration；Set iterations as M；Set i-th section of EEG signals characteristic vector w_iThe probability belonging to epileptic electroencephalogram (eeg) is p₀(y_i=1 | w_i)=0.5, i=1,2 ..., N；Set i-th section of EEG signals characteristic vector w_iPreliminary classification device be F₀(w_i)=0, i=1,2 ..., N；N=900, M=180；

2) m represents iterative steps, proceeds by following loop iteration from m=1:

I. grader F is calculated_mThe first derivative of likelihood function

${\tilde{y}}_i=2(y_i-p_{m-1}(y_i=1|w_i\left)\right),i=1,2,...,N$

Wherein p_m-1(y_i=1 | w_i) represent characteristic vector w after m-1 step iteration_iBelong to the probit of epileptic electroencephalogram (eeg)；

Ii. by method of least square by w_iRight ${\tilde{y}}_i$ Matching obtains regression coefficient r, with f (w_i) represent i-th section of EEG signals characteristic vector w_iWeak Classifier:

f(w_i)=r^Τw_i, i=1,2 ..., N；

Iii. the Weak Classifier f selected after obtaining m iteration_m

$f_m=\arg \underset{f}{min}{\underset{i=1}{\overset{N}{\Σ}}({\tilde{y}}_i-f(w_i\left)\right)}^2;$

Iv. the Bernoulli Jacob regression function L (F released by training data_m；W, Y) can be expressed as:

$L(F_m;W,Y)=\log \left(\underset{i=1}{\overset{N}{\Π}}{p}_m{\left(y_i=1|w_i\right)}^{y_i}{p}_m{\left(y_i=0|w_i\right)}^{1-y_i}\right),i=1,2,...,N;$

V. Weak Classifier weight coefficient γ after m step is calculated_mFor

${\mathrm{\γ}}_m=\arg \underset{\mathrm{\γ}}{max}L(F_{m-1}+{\mathrm{\γf}}_m;W,Y);$

Vi. grader is updated

F_m=F_m-1+εγ_mf_m

Wherein ε is a minimum value, ε=0.05；

Vii. by grader F_mCalculate characteristic vector w_iBelong to abnormal brain electricity probit:

$p_m(y_i=1|w_i)=\frac{e^{F_m\left(w_i\right)}}{e^{F_m\left(w_i\right)}+e^{-F_m\left(w_i\right)}},i=1,2,...,N$

Wherein, F_m(w_i) represent correspondence training data w after m step_iGrader；

Viii. making m=m+1, repeat above-mentioned circulation, if m=M, loop iteration terminates, and obtains grader F=F_M。

Use the characteristic vector of the normal brain activity signal of telecommunication and epileptic EEG Signal, obtain grader F by training, for power spectral density characteristic vector w of one section of K the test eeg data that leads, calculate this section of brain electricity and belong to the grader output probability p of epilepsy:

$p=\frac{e^{F\left(w\right)}}{e^{F\left(w\right)}+e^{-F\left(w\right)}}.$

Described post-processing module includes Kalman filtering and threshold decision, compares with threshold value and labelling, threshold value=0 after classification output probability p is carried out Kalman filtering, if greater than threshold value, then it is labeled as 1 (epileptic electroencephalogram (eeg)), less than threshold value, is then labeled as 0 (normal brain activity electricity).

Utilizing this invention method, the EEG signals of 20 example epileptics is carried out epilepsy detection test, obtaining correct discrimination is 98.30%.

According to Fig. 3, a kind of based on Stockwell conversion epileptic electroencephalogram (eeg) detection device disclosed in the embodiment of the present invention, wherein Fig. 3 a is the output of embodiment grader, Fig. 3 b is the result after Kalman filtering, Fig. 3 c is the result (wherein 0 represents normal brain activity electricity, and 1 represents epileptic electroencephalogram (eeg)) after threshold decision.

Pass through technique scheme, technical solution of the present invention provides the benefit that: Stockwell conversion has higher resolution compared to other Time-Frequency Analysis Method such as wavelet transformation, Short Time Fourier Transform etc., therefore the feature of extraction is more accurate after Stockwell converts, additionally, Stockwell transform operation is simple and quick and is easily achieved；The present invention can meet the demand of epileptic EEG Signal on-line checking, has good real-time and Detection results.

Claims (6)

1. an epileptic electroencephalogram (eeg) detection device based on Stockwell conversion, it is characterized in that, including eeg amplifier, data collecting card and computer, in described computer in be provided with EEG signals pretreatment module, EEG signals segmentation module, Stockwell transform characteristics extraction module, sort module and post-processing module；EEG signals is amplified by described eeg amplifier, is then gathered EEG signal by data collecting card and sends in computer；By computer, EEG signals is processed, pass sequentially through EEG signals pretreatment module, EEG signals segmentation module, Stockwell transform characteristics extraction module, sort module and post-processing module, EEG signals is carried out pretreatment, segmentation, feature extraction, training classification and post-processing operation.

A kind of epileptic electroencephalogram (eeg) detection device based on Stockwell conversion the most according to claim 1, it is characterised in that described eeg amplifier uses Neurofile NT eeg amplifier, and the EEG signals forming EEG signals is amplified；Described data collecting card uses 16 A/D to change data acquisition board, and frequency acquisition is 256Hz.

A kind of epileptic electroencephalogram (eeg) detection device based on Stockwell conversion the most according to claim 1, it is characterized in that, described EEG signals pretreatment module includes using band filter to be filtered EEG signals processing, frequency range is 0.5-30Hz, to filter the artefact in EEG signal and noise.

A kind of epileptic electroencephalogram (eeg) detection device based on Stockwell conversion the most according to claim 1, it is characterised in that described EEG signals segmentation module, EEG signals after Filtering Processing is carried out segmentation, the length of every section of EEG signals is equal, is 1024 points, time a length of 4s.

A kind of epileptic electroencephalogram (eeg) detection device based on Stockwell conversion the most according to claim 1, it is characterised in that described Stockwell transform characteristics extraction module, first carries out Stockwell conversion and obtains S EEG signals x (t) after segmentation_x(τ,f):

Wherein, τ is the time, and f is frequency；Then calculating EEG signals power spectral density feature w after Stockwell converts:

W=E{ | S_x(τ,f)|²}

Here, E{ } it is by mathematic expectaion operation.

A kind of epileptic electroencephalogram (eeg) detection dress based on Stockwell conversion the most according to claim 1.