Feature Extraction and Electrophysiological Modeling in Personalized Deep Brain Structure Using Electroencephalography Signal

Abstract

The knowledge attained from deep brain structure allows for the establishment of novel computational-based methods for yielding insights into pathophysiology and electrophysiology of neurological and degenerative brain disorders for aiding medical treatment and diagnosis, namely Epilepsy, Parkinson’s disease, and Alzheimer’s disease. To support those purposes, Electroencephalography is a readily accessible, noninvasive and relatively inexpensive monitoring method to serve diagnostic purposes, analytical tools for brain investigation, and inherently features an excellent temporal resolution for recording behaviors of brain’s spontaneous electrical activities over a period of time which follow a non-stationary dynamic process. In fact, the representation of the system also differs across individuals due to the identification of person-specific parameters. This paper proposes a scheme that extracts features characterizing the personalized spatial-temporal structure contained in EEG associated with Parkinson’s disease and Epilepsy, then build an electrophysiological model using only EEG signal for the individual to enhance accuracy for detection of these diseases. The primary result of classifying some non-linear features (Approximate entropy, Sample entropy, Fractal Dimension, and Hurst exponent) between normal and epilepsy segments showed high performance in classifiers (Decision tree: 80.5 ± 1.24%, K-nearest neighbor: 81.6 ± 1.52%, Support vector machine: 82.6 ± 0.34%).

Linh Luu and Phong Pham—Contributed equally to this work.