Electroencephalogram Feature Detection and Classification in People with Internet Addiction Disorder with Visual Oddball Paradigm (2015)

Authors: Ling, Zou; Yue, Chen; Wenjie, Li; Fan, Jing

Source: Journal of Medical Imaging and Health Informatics, Volume 5, Number 7, November 2015, pp. 1499-1503(5)

Publisher: American Scientific Publishers

Abstract:

In this paper, the electroencephalogram (EEG) signals were recorded from ten healthy and ten Internet Addiction (IA)-afflicted university students during a visual oddball paradigm. First, the original signals were preprocessed to remove some artifacts using Independent Component Analysis (ICA) algorithm. Then, the Principal Component Analysis (PCA) was employed to select a subset of channels that preserve most of the information compared to the full set of 64 channels. Finally, features of P300 waves were extracted from the event related potentials (ERPs) and compared across the target ERPs and non-target ERPs, as well as across the IA group and control group. The extracted features were further used to train four classifiers: Fisher Linear Discriminate Analysis (FLDA), Back Propagation (BP) Neural Network, Bayesian Classifier (BC) and Bayesian Regularization Back Propagation (BRBP) Neural Network. The active channels were located in the frontal, parietal, occipital and parietal-occipital areas for both healthy and IA-afflicted university students. The latency of 42 trials’ averaged ERPs under target stimulation was longer than that of 558 trials’ averaged ERPs under non-target stimulation (p < 0.05), and the amplitude of 42 trials’ averaged ERPs under target stimulation was larger than that of 558 trials’ averaged ERPs under non-target (p < 0.05). It showed significant difference in P300 amplitudes between healthy subjects and Internet Addition subjects. The amplitudes of Internet Addition were lower (p < 0.05). The classification accuracy could reach above 93% using Bayesian-based method in active areas, while it was lower than 90% in central areas. The results show that there are negative influences on the brain response and memory abilities of IA-afflicted university students. The paper deals with practical digital filter implementation to suppress 50 Hz power noise using integer coefficients filters. Very fast and simple solution enables to suppress both basic and harmonic components of power noise with nonlinear distortions. Real ECG signals were used to test effectiveness of power noise suppression. Accuracy is evaluated for basic sinusoidal and rectangular wave of noise.

Keywords: CHANNEL SELECTION; EVENT-RELATED POTENTIALS; INDEPENDENT COMPONENT ANALYSIS; P300; PATTERN RECOGNITION

Document Type: Research Article

DOI: https://doi.org/10.1166/jmihi.2015.1570

Publication date: November 1, 2015