A Study Based on P300 Component in Single-Trials for Discriminating Depression from Normal Controls

The investigation of attentional bias of depression based on P300 component has drawn interest within the last decades. Follow-up of previous research suggested the differential amplitudes between depression and normal controls (NCs) in response to various facial stimuli. In this paper, we used single-trials features in the occurrence of P300 to recognize depression from NCs. EEG activity was recorded from 24 patients and 29 NCs in a dot-probe task. We considered two traditionally used feature-extraction methods: ReliefF and principal component analysis (PCA). Then, the k-nearest neighbor (KNN), BFTree, C4.5, logistic regression and NaiveBayes were adopted in this study to make a performance comparison. The combination of NaiveBayes and PCA was applied to classify the P300 component evoked by sad-neutral pairs, which achieved higher classification accuracy than other classifiers. The classification accuracy was 98%. The classification results support that the P300 component of ERPs may reflect information processing of the specific response of depression to specific stimuli and may be applied as a physiologic index for aided diagnosis of depression in future research.