Automated EEG Analysis for Early Diagnosis of Epilepsy: A Comparative Study to Determine Relative Accuracy of Arithmetic and Huffman Coding Algorithms

Abstract

Epilepsy is a prevalent neurological disorder typically characterized by recurrent seizure activity and detected using an electroencephalogram (EEG). The manual inspection of EEG however is a challenging and slow process that is susceptive to visual errors and variability amongst subjects. Hence, significant efforts have been made towards developing algorithms for automated epilepsy diagnosis and detection. The present study focuses on comparing two algorithms employing arithmetic encoding and Huffman encoding to separate epileptic signals from seizure-free (normal) samples. The proposed diagnostic technique comprises three major steps. In the first step, discrete wavelet transform (DWT) is used to decompose the EEG signal into detail and approximation coefficients. The second step involves computation of compression ratios using encoding techniques to convert the significant coefficients into bitstreams. Finally, the compression vector set is normalized and fed to a machine learning classifier that identifies seizure activity from normal, seizure free signals. The study utilizes the standard database for epilepsy as provided by the University of Bonn in order to validate the results against prior benchmarks. The proposed methodology with arithmetic encoding algorithm achieved 100% accuracy and the classification results vary from 30.6% to 100% respectively in case of Huffman encoding. Hence, a computer aided diagnostic (CAD) technique employing DWT along with arithmetic encoding and machine learning algorithms would form a robust diagnostic system in early-stage epilepsy diagnosis.