Classification of EEG signals containing familiar and unfamiliar face stimulus using deep learning

People are skilled at facial recognition and can easily recognize a face in their brain. Distinguishing familiar and unfamiliar faces can be used as a research subject in some mental illnesses, security operations and criminal investigations. The neural changes underlying recognizing a familiar face have been the focus of scientists. These neural changes can be monitored with Electroencephalography (EEG) signals. EEG is an electrical activity of the brain recorded with the help of electrodes from the scalp. Event Related Potentials (ERP) are responses occurring within EEG signals to a stimulus. ERPs occur in the brain during face recognition. In this study, it was aimed to classify the potentials in EEG signals against familiar and unfamiliar face stimuli with high accuracy. Two paradigms were proposed to be demonstrated to 35 healthy participants. Familiar and unfamiliar face stimulus images were shown for 1000 ms in paradigm-1 and 500 ms in paradigm-2. Re-referencing the recorded EEG signals with the average reference method, filtering the EEG signals, and dividing the data into period intervals were applied respectively. Graphs of the obtained familiar and unfamiliar face stimulus data were drawn and ERPs were examined, and N250 potentials were observed in O1 and O2 channels. Bidirectional Long Short Term Memory Network (Bi-LSTM) which is one of the deep learning algorithms, was used to classify the obtained N250 potentials. In the classification step, the data set is divided into 80% training data and 20% test data. When the results were analyzed, the O2 channel achieved the highest accuracy rate with a classification performance rate of 95.6% in paradigm-1 and 96.7% in paradigm-2.