Esen Yildirim

Patient specific seizure prediction system using Hilbert spectrum and Bayesian networks classifiers.

The aim of this paper is to develop an automated system for epileptic seizure prediction from intracranial EEG signals based on Hilbert-Huang transform (HHT) and Bayesian classifiers. Proposed system includes decomposition of the signals into intrinsic mode functions for obtaining features and use of Bayesian networks with correlation based feature selection for binary classification of preictal and interictal recordings. The system was trained and tested on Freiburg EEG database. 58 hours of preictal data, 40-minute data blocks prior to each of 87 seizures collected from 21 patients, and 503.1 hours of interictal data were examined resulting in 96.55% sensitivity with 0.21 false alarms per hour, 13.896% average proportion of time spent in warning, and 33.21 minutes of average detection latency using 30-second EEG segments with 50% overlap and a simple postprocessing technique resulting in a decision (a seizure is expected/not expected) every 5 minutes. High sensitivity and low false positive rate with reasonable detection latency show that HHT based features are acceptable for patient specific seizure prediction from intracranial EEG data. Time spent for testing an EEG segment was 4.1451 seconds on average, which makes the system viable for use in real-time seizure control systems.

Patient specific seizure prediction algorithm using Hilbert-Huang Transform

Epilepsy is a neurological disorder that affects about 50 million people around the world. EEG signal processing plays an important role in detection and prediction of epileptic seizures. The aim of this study is to develop a method for early seizure prediction based on Hilbert-Huang Transform. In this patient specific method, EEG signals are decomposed into Intrinsic Mode Functions (IMFs) and first 5 IMFs are used to obtain features for classification of preictal and interictal recordings. Proposed method was tested on Freiburg EEG database. A total of 58 hours of preictal data, prior to 87 seizures, and 490 hours of interictal data were examined. Algorithm resulted in 89.66% sensitivity (78 of 87 seizures) and 0.49 FPs/h using 30 seconds EEG segment with 50% overlap.

Emotion primitives estimation from EEG signals using Hilbert Huang Transform

This paper addresses the problem of emotion primitives estimation using information obtained from EEG signals. The EEG data were collected from 18 subjects, 9 male and 9 female, aged from 19 to 26 years old. We used audio clips from International Affective Digital Sounds (IADS) as stimuli for emotion elicitation. Hilbert-Huang Transform, a proper method for non-linear and non-stationary signal processing, was used for feature extraction. EEG signals were first decomposed into their Intrinsic Mode Functions (IMFs). Then 990 features were computed from the first five IMFs. To identify the most salient features and eliminate the redundant and irrelevant ones, we performed correlation based feature selection (CFS). This feature selection process reduced the number of features dramatically while increasing the performance remarkably. In this work, we used support vector regression for estimation of each emotion primitive value. Regression mean absolute error values and their standard deviations over all subjects for valence, activation, and dominance were obtained as 1.11 (0.13), 0.65 (0.09) and 0.38 (0.06) respectively.

Emotion estimation from EEG signals using wavelet transform analysis

Emotion recognition is important for an effective human-machine interaction. Information obtained from speech, gestures and mimics, heart rate, and temperature can be used in emotion estimation. In this study, emotion estimation from EEG signals using wavelet decomposition is performed. For this purpose, EEG signals were recorded from 20 subjects and audio stimuli are used to evoke emotions. Delta, Theta, Alfa, Beta and Gamma sub-bands of signals are computed using wavelet transform. Statistical features and energy of each band are computed. Correlation based feature selection algorithm is applied to the base feature set to obtain the most relevant subset and emotion primitives are estimated using Support Vector Regression. Emotion estimation results in terms of mean absolute error using db4, db8 and coif5 mother wavelets are 0.28, 0.26, and 0.29 for valence, 0.20, 0.20, and 0.19 for activation and 0.11, 0.10, and 0.10 for dominance respectively.

Comparison of wavelets for classification of cognitive EEG signals

In this work, different wavelet types, that have been frequently used in EEG signal analysis and classification, are compared for cognitive EEG classification. EEG signals are collected from 18 healthy subjects during math processing and simple text reading. Symlet, coiflet and bior wavelet types are used for feature extraction and classification performances of BayesNet and J48 classifiers are compared. The best true positive rate of 90.6% is obtained using Boir 2.4 wavelet type with J48 classifier.

Orthogonal Extreme Learning Machine Based P300 Visual Event-Related BCI

Brain Computer Interface BCI is a type of human-computer relationship research that directly translates electrical activity of brain into commands that can rule equipment and create novel communication channel for muscular disabled patients. In this study, in order to overcome shortcoming of Singular Value Decomposition in Extreme Learning Machine, iteratively optimized neuron numbered QR Decomposition technique with different approaches are proposed. QR Decomposition Extreme Learning Machine technique based P300 event-related potential BCI application that achieves almost % 100 classification accuracy with milliseconds is presented. QR decomposition based ELM and novel feature extraction method named Multi Order Difference Plot MoDP techniques are milestones of proposed BCI system.

Patient Specific Congestive Heart Failure Detection From Raw ECG signal

In this study; in order to diagnose congestive heart failure (CHF) patients, non-linear second-order difference plot (SODP) obtained from raw 256 Hz sampled frequency and windowed record with different time of ECG records are used. All of the data rows are labelled with their belongings to classify much more realistically. SODPs are divided into different radius of quadrant regions and numbers of the points fall in the quadrants are computed in order to extract feature vectors. Fishers linear discriminant, Naive Bayes, Radial basis function, and artificial neural network are used as classifier. The results are considered in two step validation methods as general k-fold cross-validation and patient based cross-validation. As a result, it is shown that using neural network classifier with features obtained from SODP, the constructed system could distinguish normal and CHF patients with 100% accuracy rate.

Emotion recognition from speech using Fisher's discriminant analysis and Bayesian classifier

In this study, a large number of features that were obtained to classify speech emotions were projected into different spaces, selecting different numbers of principal components in principal component analysis and Fishers discriminant analysis. Classifications were performed in those spaces using Naïve-Bayes classifier and obtained results were compared. While the highest accuracy obtained in the Fisher space was 57.87%, it was calculated as 48.02% in the principal component space.

Classification of emotional valence dimension using artificial neural networks

Emotions play an important role in human interaction. Emotion recognition should be considered to design an effective Brain-Computer Interface. In this work binary classification (low/high) for valence which is one of the primitives used in expressing emotions is performed. Hilbert-Huang Transform is used for feature extraction, multi layer feed forward Artificial Neural Networks is used for subject independent classification and 69% of true positive rate is obtained.

Classification of emotion primitives from EEG signals using visual and audio stimuli

Emotion recognition from EEG signals has an important role in designing Brain-Computer Interface. This paper compares effects of audio and visual stimuli, used for collecting emotional EEG signals, on emotion classification performance. For this purpose EEG data from 25 subjects are collected and binary classification (low/high) for valence and activation emotion dimensions are performed. Wavelet transform is used for feature extraction and 3 classifiers are used for classification. True positive rates of 71.7% and 78.5% are obtained using audio and video stimuli for valence dimension 71% and 82% are obtained using audio and video stimuli for arousal dimension, respectively.