Epileptic Seizure Detection using Bidimensional Empirical Mode Decomposition and Distance Metric Learning on Scalogram

Epileptic seizure detection through visual inspection of Electroencephalogram (EEG) signals is a tedious task demanding high level expertise as well as time. Automatic seizure detection is one of the solutions suggested by engineering researchers working in the field of biomedical signal processing. In classical approach, EEG signals are first preprocessed to apply signal processing algorithm for feature extraction and then classified for seizure detection. The efficiency of the classifier is highly dependent on the discriminative space, so, the challenge in the proposed approach is to extract features and to apply classifier efficiently so that seizures may be detected well. The two dimensional visual representation of EEG signals (scalogram) obtained through Continuous Wavelet transform is utilized for feature extraction. Local Binary Pattern (LBP) and Histogram of Oriented Gradients (HOG) of instantaneous frequency components of the scalograms are calculated as potential feature values. The extracted features are mapped to discriminative space using various distance metric learning algorithm and mapped features are fed to Support Vector Machine for classification. The proposed algorithm is evaluated on the EEG database to prove its efficacy. The results indicate superior performance of Laplacian Eigenmaps method for dimensionality reduction with 99.08 % classification accuracy. The proposed methodology is novel and outperforms the state-of-the-art methods of epileptic seizure detection.