Signal Selection Technique based on Statistical Approach for Enhanced Detection of Single-trial Auditory Evoked Potentials

Volume conduction from insignificant neuronal sources in the brain poses a challenge to the detection and classification of single-trial low amplitude evoked potentials in electroencephalograph (EEG). This work presents a statistical signal selection method for enhanced detection of single-trial EEG auditory evoked potential (AEP) elicited in the brain in response to subjects' own name audio stimulus. The proposed method comprises of a signal selection stage based on a statistical analysis followed by a support vector machine (SVM)-based classifier. The EEG signals recorded from the Fp1 electrode of 24 subjects are used to generate a classifier-dependent feature vector. With the selected one-quarter of AEP signals, a single-trial classification accuracy of 70.59% is obtained. To the best of our knowledge, this is the first study that reports the classification of single-trial AEPs evoked by subjects' own-name audio stimulus versus familiar-name audio stimulus.