Ffeature extraction of epilepsy eeg using discrete wavelet transform
Download as PPTX, PDF3 likes2,041 views
This document summarizes a presentation on feature extraction of epilepsy EEG signals using discrete wavelet transforms. The presentation discusses EEG data acquisition from public datasets containing healthy and epileptic patient recordings. It then describes using discrete wavelet transforms to decompose EEG signals into different frequency sub-bands, and extracting statistical features from each sub-band like maximum, minimum, mean, standard deviation, and entropy. These extracted features are used to classify EEG signals as normal or epileptic. The approach decomposes signals into 5 sub-bands corresponding to delta, theta, alpha, beta, and gamma frequency ranges to capture characteristics of different brain states for epilepsy identification.
![Related Work
Ref. Method # of Features Remark
[8] EMD Eight features The EMD generates the set of amplitude and frequency modulated components known as intrinsic mode functions
(IMFs).
Two area measures have been computed, one for the graph obtained as the analytic signal representation of IMFs
in the complex plane and another for second-order difference plot (SODP) of IMFs of EEG signals. Both of these
area measures have been computed for first four IMFs of the normal and epileptic seizure EEG signals. These eight
features obtained from both area measures of first four IMFs have been used as input feature set for classification
of normal and epileptic seizure EEG signals using least square support vector machine (LS-SVM) classifier.
[9] DWT Four features The proposed technique involved training the ANFIS classifier to detect epileptic seizure in EEG when the statistical
features extracted from the wavelet sub-bands of EEG signals were used as inputs
[10] DWT Four features Proposed a wavelet-based feature extraction technique which consequently uses simple statistical parameters to
detect epileptic EEG signals using a back propagating artificial neural network classifier.
[11] DWT One feature Approximate Entropy (ApEn) for epilepsy detection from EEG signals is used.
[12] ICA Three features Improving the accuracy of EEG signal classification is presented to detect epileptic seizures. ICA is incorporated as
a preprocessing step and Short-Time Fourier Transform (STFT) is used for de-noising the signal adequately.
[13] ICA four features The signals were decomposed into the frequency sub-bands using DWT and a set of statistical features was
extracted from the subbands to represent the distribution of wavelet coefficients. Principal components analysis
(PCA), independent components analysis (ICA) and linear discriminant analysis (LDA) is used to reduce the
dimension of data. Then these features were used as an input to a support vector machine (SVM) with two discrete
outputs: epileptic seizure or not.
[14] DWT Four features Original EEG signals representation by wavelet packet coefficients and feature extraction using the best basis-
based wavelet packet entropy method.
[15] DWT Four features Each EEG signal is decomposed into five constituent EEG sub-bands by DWT. The nonlinear parameters of each
sub-band and the original EEG are quantified in the form of the time lag (TL), the embedding dimension (ED), the
correlation dimension (CD), and the largest Lyapunov exponent (LLE).
7
fewer previous research on the feature extraction methods in EEG signals](https://image.slidesharecdn.com/revisedabofeatureextractionofepilepsyeegusingdiscretewavelettransform-161228153603/85/Ffeature-extraction-of-epilepsy-eeg-using-discrete-wavelet-transform-7-320.jpg)
Ffeature extraction of epilepsy eeg using discrete wavelet transform
- 1. Feature Extraction of Epilepsy EEG using Discrete Wavelet Transform Asmaa Hamad Elsaied ICENCO Cairo 2016, Cairo University (29-December-2016) Master Student, Faculty of Computers and Information, Mina University
- 2. Introduction Related Work Materials and Methods EEG Data Acquisition Discrete wavelet transforms (DWT) Proposed Approach Results and Discussion Conclusion and future work 2 Agenda ICENCO Cairo 2016
- 3. Introduction Epilepsy is one of the most common a chronic neurological disorders of the brain that affect millions of the world’s populations. It is characterized by recurrent seizures, which are physical reactions to sudden, usually brief, excessive electrical discharges in a group of brain cells. Hence, seizure identification has great importance in clinical therapy of epileptic patients. Electroencephalogram (EEG) is most commonly used in epilepsy detection since it includes precious physiological information of the brain. 3 ICENCO Cairo 2016
- 4. Introduction Electroencephalogram (EEG) The EEG signal is usually used for the purpose of recording the electrical activities of the brain signal that typically arises in the human brain. The recording of the electrical activity is basically done by placing electrodes on the scalp, which measures the voltage fluctuations in the brain. 4 ICENCO Cairo 2016 EEG contains lots of worthy information relating to the numerous physiological states of the brain and thus is a very useful tool for understanding the brain disease, such as epilepsy. EEG signals of epileptic patients exhibit two states of abnormal activities namely interictal or seizure free (in-between epileptic seizures) and ictal (in the course of an epileptic seizure).
- 5. Introduction Electroencephalogram (EEG) sub-bands The EEG signals are commonly decomposed into five EEG sub- bands: delta, theta, alpha, beta and gamma. 5 ICENCO Cairo 2016
- 6. Introduction Electroencephalogram (EEG) Frequency range and amplitude for each type of waves 6 Wave Frequency range Amplitude Delta band 0.5 – 4 Hz High Theta band 4 – 8 Hz Low-medium Alpha band 8 – 15 Hz Low Beta band 15 – 30 Hz Very low Gamma band 30 – 60 Hz Smallest ICENCO Cairo 2016
- 7. Related Work Ref. Method # of Features Remark [8] EMD Eight features The EMD generates the set of amplitude and frequency modulated components known as intrinsic mode functions (IMFs). Two area measures have been computed, one for the graph obtained as the analytic signal representation of IMFs in the complex plane and another for second-order difference plot (SODP) of IMFs of EEG signals. Both of these area measures have been computed for first four IMFs of the normal and epileptic seizure EEG signals. These eight features obtained from both area measures of first four IMFs have been used as input feature set for classification of normal and epileptic seizure EEG signals using least square support vector machine (LS-SVM) classifier. [9] DWT Four features The proposed technique involved training the ANFIS classifier to detect epileptic seizure in EEG when the statistical features extracted from the wavelet sub-bands of EEG signals were used as inputs [10] DWT Four features Proposed a wavelet-based feature extraction technique which consequently uses simple statistical parameters to detect epileptic EEG signals using a back propagating artificial neural network classifier. [11] DWT One feature Approximate Entropy (ApEn) for epilepsy detection from EEG signals is used. [12] ICA Three features Improving the accuracy of EEG signal classification is presented to detect epileptic seizures. ICA is incorporated as a preprocessing step and Short-Time Fourier Transform (STFT) is used for de-noising the signal adequately. [13] ICA four features The signals were decomposed into the frequency sub-bands using DWT and a set of statistical features was extracted from the subbands to represent the distribution of wavelet coefficients. Principal components analysis (PCA), independent components analysis (ICA) and linear discriminant analysis (LDA) is used to reduce the dimension of data. Then these features were used as an input to a support vector machine (SVM) with two discrete outputs: epileptic seizure or not. [14] DWT Four features Original EEG signals representation by wavelet packet coefficients and feature extraction using the best basis- based wavelet packet entropy method. [15] DWT Four features Each EEG signal is decomposed into five constituent EEG sub-bands by DWT. The nonlinear parameters of each sub-band and the original EEG are quantified in the form of the time lag (TL), the embedding dimension (ED), the correlation dimension (CD), and the largest Lyapunov exponent (LLE). 7 fewer previous research on the feature extraction methods in EEG signals
- 8. EEG Data Sets The experimental data used is publically available Bonn data set “Klinik für Epileptologie, Universität Bonn’’. The dataset includes five different sets: 8 ICENCO Cairo 2016 • 5 awake healthy subjects with eyes open • Surface EEG recordingA • 5 awake healthy subjects with eyes closed • Surface EEG recordingB • Inter-ictal EEG from five epileptic patients • intracranial depth electrodes from hippocampal formation of opposite hemisphere the brain C • Inter-ictal EEG from five epileptic patients • Intracranial depth electrodes from epileptogenic zone.D • Ictal EEG from five epileptic patients • depth and strip electrodesE
- 9. EEG Data Samples EEG signals of each dataset. 9 ICENCO Cairo 2016
- 10. The EEG signals of three subsets namely A, D, and E have been used. 10 ICENCO Cairo 2016 EEG Data Sets Characterstics Settings Set A Set D Set E Subjects 5 healthy 5 epileptic patients 5 epileptic patients Electrode type surface Intracranial Intracranial Electrode placement International 10- 20 system Hippocampal formation Epileptogenic zone Patient’s state Awake, eyes open Seizure-free (Interictal) Seizure activity (Ictal) Number of epochs 100 100 100 Epoch duration (s) 23.6 23.6 23.6
- 11. Discrete wavelet transforms (DWT) A wavelet is a short wave, which has its energy intensified in time to give a tool for the analysis of transient, non- stationary signals or time-varying phenomena . If a signal does not change much over time, we would call it as a stationary signal. Fourier transform could be applied to the stationary signals easily and a good result can be taken. 11 ICENCO Cairo 2016
- 12. Discrete wavelet transforms (DWT) However, many signals like EEG having the non- stationary and transient characteristics, in such situation ideally Fourier transform may not be applied directly. But time–frequency methods can be used . Wavelet transform method has been used to extract the individual EEG sub-bands and reconstruct the information accurately because the wavelet transform has the advantages of: time-frequency localization, multi-rate filtering, and scale-space analysis. 12 ICENCO Cairo 2016
- 13. The Proposed Approach (General model) Identify the epileptic seizure 13 ICENCO Cairo 2016
- 14. The Proposed Approach Feature Extraction using DWT Each EEG signal is decomposed into five constituent EEG sub-bands by discrete wavelet transform (DWT). The EEG epochs were analyzed into various frequency bands by using fourth-order Daubechies (db4) wavelet function up to 4th- level of the decomposition. The statistical parameter like entropy, min, max, mean, median, standard deviation, variance, energy and Relative Wave Energy (RWE) were computed for feature extraction. 14 ICENCO Cairo 2016
- 15. The Proposed Approach Feature Extraction using DWT Feature extraction is a special form of dimensionality reduction. When the input data to an algorithm is too large to be processed and it is suspected to be notoriously redundant (much data, but not much information) then the input data will be transformed into a reduced representation set of features (also named features vector). Transforming the input data into the set of features is called feature extraction. If the features extracted are carefully chosen it is expected that the features set will extract the relevant information from the input data in order to perform the desired task using this reduced representation instead of the full size input. 15 ICENCO Cairo 2016
- 16. Following features of wavelet coefficients from each sub band that were extracted to classify EEG signals. Maximum of the wavelet coefficients in each sub-band. Minimum of the wavelet coefficients in each sub-band. Mean of the wavelet coefficients in each sub-band The standard deviation of the wavelet coefficients in each sub-band. The variance of the wavelet coefficients in each sub- band is the square of the standard deviation. The median of the wavelet coefficients in each sub- band. 16 ICENCO Cairo 2016 The Proposed Approach Feature Extraction using DWT Skewness of the wavelet coefficients in each sub-band. A measure of the asymmetry of the data distribution. Energy in the sub-band The energy points out that the strength of the signal as it gives the area under the curve of power at any interval of time. Relative Wave Energy (RWE) in the sub-band RWE characterize the relative energy in each frequency sub-band and is utilize to detect the correspondence between segments of EEG signal.
- 17. Entropy in the sub-band. Entropy is a numerical measure of uncertainty (doubt) of outcome where signal contained thousands of bits of information. Based on the above mentioned, ten features were extracted for chosen categories of signals to create the original feature database at each decomposition level starting from D1–D4 and one final approximation, A4. These are extracted to help in distinguishing between normal and epileptic signal. 17 ICENCO Cairo 2016 The Proposed Approach Feature Extraction using DWT
- 18. Decomposition level Sub-band signal Frequency band (Hz) 1 D1(gamma) 30-60 2 D2 (beta) 15-30 3 D3 (alpha) 8-15 4 D4 (theta) 4-8 4 A4 (delta) 0-4 18 ICENCO Cairo 2016 Experimental Results
- 19. 19 Approximate and coefficients are taken from a healthy subject (set A). ICENCO Cairo 2016 Experimental Results
- 20. 20 Approximate and coefficients are taken from an epileptic subject (set D). ICENCO Cairo 2016 Experimental Results
- 21. 21 Approximate and coefficients are taken from an epileptic subject (set E). ICENCO Cairo 2016 Experimental Results
- 22. 22 EXTRACTED FEATURE COEFFICIENTS FOR THE LAST EPOCH OF SET A Set Features Wavelet Sub-bands D1 (Gamma) D2 (Beta) D3 (Alpha) D4 (Theta) A4 (Delta) Set A Max Min Mean Median Variance Deviation Energy RWE Entropy Skewness 9.946874e-04 -9.265823e-04 -1.269325e-07 3.465760e-06 6.455155e-08 2.540700e-04 1.323953e-04 8.794600e-06 -2.087377e-03 3.727474e-01 7.879125e-03 -8.628986e-03 -1.483079e-06 -4.923745e-05 5.112694e-06 2.261127e-03 5.255851e-03 3.491297e-04 -5.988027e-02 4.730014e-01 3.901077e-02 -4.608552e-02 -1.217418e-04 -1.900916e-04 1.369548e-04 1.170277e-02 7.081328e-02 4.703904e-03 -5.705388e-0 7.493647e-01 7.471598e-02 -7.253425e-02 1.033281e-03 -5.439984e-04 6.086433e-04 2.467070e-02 1.591356e-01 1.057088e-02 -1.056781e+00 2.714428e-02 6.104268e-01 -8.788146e-01 -7.732445e-03 -1.336441e-02 5.671705e-02 2.381534e-01 1.481882e+01 9.843673e-01 -2.977001e+01 6.492470e-01 ICENCO Cairo 2016 Experimental Results
- 23. 23 EXTRACTED FEATURE COEFFICIENTS FOR THE LAST EPOCH OF SET D Set Extracte d Feature s Wavelet Sub-bands D1 (Gamma) D2 (Beta) D3 (Alpha) D4 (Theta) A4 (Delta) Set D Max Min Mean Median Variance Deviation Energy RWE Entropy Skewness 3.969254e-04 -3.742630e-04 -1.062821e-07 -3.011566e-06 7.481376e-09 8.649495e-05 1.534432e-05 2.407420e-06 -2.750434e-04 4.457542e-01 4.264739e-03 -4.189743e-03 2.729993e-06 2.262996e-05 5.425604e-07 7.365870e-04 5.577597e-04 8.750869e-05 -7.531945e-03 2.173789e+00 1.840372e-02 -2.153365e-02 -5.010535e-05 -4.831392e-05 2.228588e-05 4.720792e-03 1.152310e-02 1.807896e-03 -1.118578e-01 2.088849e+00 7.366416e-02 -8.237293e-02 1.173667e-04 5.086258e-04 5.202897e-04 2.280986e-02 1.357992e-01 2.130597e-02 -8.970721e-01 1.364383e+00 3.685789e-01 -4.581400e-01 -7.577910e-03 -1.029943e-02 2.379626e-02 1.542604e-01 6.225868e+00 9.767962e-01 - 1.877158e+01 -2.049270e-01 ICENCO Cairo 2016 Experimental Results
- 24. 24 EXTRACTED FEATURE COEFFICIENTS FOR THE LAST EPOCH OF SET E Set Extracte d Feature s Wavelet Sub-bands D1 (Gamma) D2 (Beta) D3 (Alpha) D4 (Theta) A4 (Delta) Set E Max Min Mean Median Variance Deviation Energy RWE Entropy Skewness 3.955192e-03 -3.549488e-03 6.948730e-07 -2.612595e-05 7.854200e-07 8.862392e-04 1.610897e-03 1.645268e-05 -2.096376e-02 1.662602e+00 3.189520e-02 -2.833828e-02 -5.382679e-06 -3.217398e-04 9.073279e-05 9.525376e-03 9.327334e-02 9.526346e-04 -7.912125e-01 2.449215e-01 1.880762e-01 -1.478077e-01 1.642593e-03 -4.622447e-03 3.228821e-03 5.682271e-02 1.670698e+00 1.706345e-02 -8.436742e+00 -2.552237e-01 5.564677e-01 -7.536238e-01 1.904032e-04 -9.030940e-03 7.326989e-02 2.706841e-01 1.912345e+01 1.953148e-01 - 4.057527e+01 -9.135021e-01 2.277090e+00 - 2.324888e+00 3.123408e-03 1.971368e-03 2.950933e-01 5.432249e-01 7.702190e+01 7.866527e-01 - 8.908658e+00 2.659280e+00ICENCO Cairo 2016 Experimental Results
- 25. Conclusion An improved DWT technique to extract ten features from EEG signal in which can be used to classify epileptic seizure. The number of features extracted using the improved DWT has considered the best when compared to other studies as have demonstrated in related work section. 25 ICENCO Cairo 2016
- 26. Future work We plan to select the significant and relevant features from these huge number of features based on swarm optimization algorithms like Whale Optimization Algorithm (WOA). then ANN is used for the classification, which it can be easily distinguished between normal and epileptic. 26 ICENCO Cairo 2016
- 27. Thanks and Acknowledgement 27 ICENCO Cairo 2016