Language Mapping using tEEG and EEG Data with Convolutional Neural Networks

Tripolar electroencephalography (tEEG) has been found to have significantly better signal-to-noise ratio, spatial resolution, mutual information, and high-frequencies compared to EEG. This paper analyzes the tEEG signals acquired simultaneously with the EEG signals and compares their ability to map language to left and right hemispheres using convolutional neural networks (CNNs). The results show that while the time-domain features of tEEG and EEG signals lead to comparable functional mapping, the frequency domain features are significantly different. The left and right hemisphere classification performances using tEEG are equivalent in time and frequency domains. However, frequency domain classification for EEG results in less accuracy. Clinical Relevance- This technique could quickly, and noninvasively, guide clinicians about language dominance when preparing for resective surgery.