Deep learning helps EEG signals predict different stages of visual processing in the human brain

Analysis of electroencephalogram (EEG) signals to determine the nature of visual stimuli, being experienced by a person, is an active area of research. It is key to understand the link between human brain and behavior, especially for brain computer interface (BCI) applications and rehabilitation of patients suffering with neurological disorders. In this research, we conducted an experiment comparing two stages of visual processing, determined distinct EEG signals associated with them, and subsequently used a classifier to distinguish the two stages. EEG data was collected using a feature-binding experiment that required subjects to detect changes in color and shape binding after 100 ms and after 1500 ms. The two stages denoted by these study-test intervals were determined using features extracted from both time and frequency domains. These were used to separately train various machine learning classifiers. The time-frequency domain representation of the signal was used to train a convolutional neural network (CNN). Promising results were obtained. Thus, the contribution of the paper is two-fold. Firstly, we carry out EEG data analysis using deep learning to classify whether the EEG trial belongs to 100 ms class or 1500 ms class. Secondly, we connect these results to predict different stages of visual processing in human brain and visual feature binding. Thus, deep learning can help us predict the stages of visual processing and, hence, unlock important insights regarding the temporal dynamics of brain functioning. This can help in building relevant tools for BCI applications such as neuro-rehabilitation of subjects suffering impairments in visual feature binding.