Use of Brain Electrical Activity to Classify People with Concussion: A Deep Learning Approach

Concussions are mild traumatic brain injuries that can interfere with the function of the brain. Electroencephalograph (EEG) recordings are used to identify certain diseases and abnormalities of the brain function. Though it is widely used due to low cost and noninvasive nature, getting reliable results from an EEG recording requires a trained professional. The EEG analysis and features extracted varies according to the situation, and the ideal methodology is still indeterminate. This paper proposes a novel deep learning classifier to differentiate between people who previously had concussions and those who did not, using minimally preprocessed EEG recordings. The model will eliminate the need for time-consuming sophisticated feature extraction methods by allowing the classifier to choose the features itself. The proposed deep learning model consists of Long Short Term Memory (LSTM) cell units that are capable of learning long term dependencies and hence widely used for sequence classification models. The proposed model shows an accuracy of 92.86% and an AUROC of 0.91.