Unsupervised Quantification of High-Gamma Activity in Electrocorticographic Signals

Abstract

Power modulation in the high-gamma frequency band of brain waves, commonly referred to as high-gamma activity, is a widely recognized neural correlate of cognitive and behavioral tasks. We here propose a novel method that robustly determines the signal-to-noise ratio of high-gamma activity estimates and thereby quantifies the amount of the underlying physiological activity. We validated our method by correlating its output to established metrics, including classification accuracy in a brain-computer interface environment and z-scores in a functional mapping scenario. We obtained correlation coefficients between 0.84 and 0.97, which demonstrates excellent agreement. All results are statistically highly significant. Unlike conventional approaches, our method is unsupervised. It can quantify task-related high-gamma activity without knowing the task itself. This enables new application scenarios in the field of brain-computer interfacing, neuroscience, and neurosurgery.