Academia.eduAcademia.edu

Figure 9 – uploaded by syed muhammad usman

See full PDFdownloadDownload figure
Comparison of seizure prediction results obtained from proposed methodology with existing methods. seizures predicted by the proposed method. It has been concluded that p is less than 0.05 in case of all subjects, therefore the proposed method performs better.

Table 9 Comparison of seizure prediction results obtained from proposed methodology with existing methods. seizures predicted by the proposed method. It has been concluded that p is less than 0.05 in case of all subjects, therefore the proposed method performs better.

Related Figures (16)

Fig. 1. Interictal, Preictal, Ictal and Postictal states of Seizures: Ictal state is between two red lines, after red lines is postictal state, between red and blue line is preictal state and before blue line is interictal state. Researchers have used multiple classifiers to classify preictal and interictal states EEG signals including k nearest neighbor Accurate prediction of epileptic seizures has always remained a challenge due to low SNR in EEG signals acquired from the scalp or brain’s surface of epilepsy affected patient, features with high interclass variance and classification of preictal and interictal classes with increased true positive rate and decreased false positive rate. In case of scalp EEG signals, problem of low SNR is very common due to inter- ference between multiple electrodes as the placement of electrodes is
Fig. 1. Interictal, Preictal, Ictal and Postictal states of Seizures: Ictal state is between two red lines, after red lines is postictal state, between red and blue line is preictal state and before blue line is interictal state. Researchers have used multiple classifiers to classify preictal and interictal states EEG signals including k nearest neighbor Accurate prediction of epileptic seizures has always remained a challenge due to low SNR in EEG signals acquired from the scalp or brain’s surface of epilepsy affected patient, features with high interclass variance and classification of preictal and interictal classes with increased true positive rate and decreased false positive rate. In case of scalp EEG signals, problem of low SNR is very common due to inter- ference between multiple electrodes as the placement of electrodes is
Existing methods for detection of preictal state using scalp EEG signals.
Existing methods for detection of preictal state using scalp EEG signals.
Fig. 2. Flow Chart of the proposed method.
Fig. 2. Flow Chart of the proposed method.
Where, f € {1,...., F} is the index of frequency bin and F € N is the size of DFT. The input sequence of DFT x;[t] has been interpolated with zero padding as F > T. X[f, I] is the extracted STFT of EEG signal frame x [n] and represents the local time frequency behavior of the signal around the time index /H and the frequency bin k. After converting time Fig. 3. Non-overlapping and overlapping window processing and the corresponding spectrograms.
Where, f € {1,...., F} is the index of frequency bin and F € N is the size of DFT. The input sequence of DFT x;[t] has been interpolated with zero padding as F > T. X[f, I] is the extracted STFT of EEG signal frame x [n] and represents the local time frequency behavior of the signal around the time index /H and the frequency bin k. After converting time Fig. 3. Non-overlapping and overlapping window processing and the corresponding spectrograms.
Fig. 4. a) 2D plot of Mean and Variance b) 2D plot of Skewness and Kurtosis. The optimal feature set obtained from PSO is then fed into three different classifiers including CNN, SVM and LSTM. Bagging classifier is then applied with the output of these classifiers to get the final class label which can be either preictal or interictal. Upon successful detection ot preictal state samples, preventive measures can be taken to partially control or stop the upcoming seizure. This can reduce the risk of injury and is beneficial for health of epilepsy affected patient. To perform classification 56 s time is required to process a new sample. If correlation between two features is more than a specific threshold say 0.5 or — 0.5 in negative side, then it shows that features are either positive or negative correlated. In either case, both these features do not add any useful information for classification between preictal and interictal state rather increases processing power and lead towards curse of dimensionality. Correlated features add redundancy and do not help in increasing classification accuracy for prediction of epileptic seizures. PCC helps in finding correlation between same features across multiple channels so that highly correlated features may be dropped to reduce the
Fig. 4. a) 2D plot of Mean and Variance b) 2D plot of Skewness and Kurtosis. The optimal feature set obtained from PSO is then fed into three different classifiers including CNN, SVM and LSTM. Bagging classifier is then applied with the output of these classifiers to get the final class label which can be either preictal or interictal. Upon successful detection ot preictal state samples, preventive measures can be taken to partially control or stop the upcoming seizure. This can reduce the risk of injury and is beneficial for health of epilepsy affected patient. To perform classification 56 s time is required to process a new sample. If correlation between two features is more than a specific threshold say 0.5 or — 0.5 in negative side, then it shows that features are either positive or negative correlated. In either case, both these features do not add any useful information for classification between preictal and interictal state rather increases processing power and lead towards curse of dimensionality. Correlated features add redundancy and do not help in increasing classification accuracy for prediction of epileptic seizures. PCC helps in finding correlation between same features across multiple channels so that highly correlated features may be dropped to reduce the
Details of all subjects participated in recordings of CHBMIT dataset. Table 3
Details of all subjects participated in recordings of CHBMIT dataset. Table 3
Number of parameters required in the propsoed CNN architecture. Table 2 4. Results and discussion
Number of parameters required in the propsoed CNN architecture. Table 2 4. Results and discussion
Evaluation of Epileptic Seizure Prediction Method with STFT and Bandpass filtering for noise removal, combined feature set having both Handcrafted and CNN based features, Pearson Correlation Coefficient for feature selection and Classification using SVM.
Evaluation of Epileptic Seizure Prediction Method with STFT and Bandpass filtering for noise removal, combined feature set having both Handcrafted and CNN based features, Pearson Correlation Coefficient for feature selection and Classification using SVM.
Fig. 5. A three layered architecture of CNN for machine learned feature extraction.
Fig. 5. A three layered architecture of CNN for machine learned feature extraction.
Evaluation of Epileptic Seizure Prediction Method with STFT and Bandpass filtering for noise removal, combined feature set having both Handcrafted and CNN based features, Pearson Correlation Coefficient and Particle Swarm Opti- mization for feature selection and Classification using SVM.
Evaluation of Epileptic Seizure Prediction Method with STFT and Bandpass filtering for noise removal, combined feature set having both Handcrafted and CNN based features, Pearson Correlation Coefficient and Particle Swarm Opti- mization for feature selection and Classification using SVM.
Evaluation of Epileptic Seizure Prediction Method with STFT and Bandpass filtering for noise removal, combined feature set having both Handcrafted and CNN based features, Pearson Correlation Coefficient and Particle Swarm Opti- mization for feature selection and Classification using CNN.
Evaluation of Epileptic Seizure Prediction Method with STFT and Bandpass filtering for noise removal, combined feature set having both Handcrafted and CNN based features, Pearson Correlation Coefficient and Particle Swarm Opti- mization for feature selection and Classification using CNN.
Evaluation of Epileptic Seizure Prediction Method with STFT and Bandpass filtering for noise removal, combined feature set having both Handcrafted and CNN based features, Pearson Correlation Coefficient and Particle Swarm Opti- mization for feature selection and Classification using LSTM.
Evaluation of Epileptic Seizure Prediction Method with STFT and Bandpass filtering for noise removal, combined feature set having both Handcrafted and CNN based features, Pearson Correlation Coefficient and Particle Swarm Opti- mization for feature selection and Classification using LSTM.
Fig. 6. Comparison of seizure prediction results obtained from proposed methodology with existing methods.
Fig. 6. Comparison of seizure prediction results obtained from proposed methodology with existing methods.
Evaluation of Epileptic Seizure Prediction Method with STFT and Bandpass filtering for noise removal, combined feature set having both Handcrafted and CNN based features, Pearson Correlation Coefficient and Particle Swarm Opti- mization for feature selection and Classification using Ensemble Classifier. anticipation time of epileptic seizure prediction method. Therefore, it is concluded that ensemble classifier gives the optimal classification per- formance compared to result achieved from individual classifiers. Schelter et al. (2006) have proposed the method to compute the prob- ability for alarm generation upon prediction of preictal state which has been used for statistical validation of the seizure onset and can be computed as follows:
Evaluation of Epileptic Seizure Prediction Method with STFT and Bandpass filtering for noise removal, combined feature set having both Handcrafted and CNN based features, Pearson Correlation Coefficient and Particle Swarm Opti- mization for feature selection and Classification using Ensemble Classifier. anticipation time of epileptic seizure prediction method. Therefore, it is concluded that ensemble classifier gives the optimal classification per- formance compared to result achieved from individual classifiers. Schelter et al. (2006) have proposed the method to compute the prob- ability for alarm generation upon prediction of preictal state which has been used for statistical validation of the seizure onset and can be computed as follows:
Related topics:
Seizure PredictionEEG Signal Analysis
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved