STLDA: A Spatiotemporal Linear Discriminant Analysis for Single-trial ERP-based Depression Recognition

Yuhan Shi, Nan Zhao, Wenjie Yuan, Qiqi Zhao, Xiaowei Zhang, Bin Hu

Abstract: Event-related potentials (ERP) within the Electroencephalogram (EEG), in particular, are electric signals induced by stimuli that can reflect specific cognitive activities of the brain. Therefore, ERP can be used as an objective biomarker to distinguish patients with depression from healthy individuals. However, the analysis and classification of single-trial ERP are difficult due to the high trial-to-trial variability and the low signal-to-noise ratio (SNR). Therefore, how to improve the SNR and the single-trial classification accuracy has received much attention. In this study, we proposed a spatiotemporal linear discriminant analysis (STLDA) method for single-trial ERP-based depression recognition, which can obtain optimal temporal and spatial filters for ERP signals to enhance their SNR significantly and preserve the spatiotemporal characteristics of the signals for depression recognition using the minimum distance to mean (MDM) classification strategy. Experimental results on two public datasets showed that our method achieved higher classification accuracy in comparison with some baseline methods. Further, the depression recognition performance of our method is almost equal to the traditional trial-averaged strategy.