Unravelling Causal Relationships Between Cortex and Muscle with Errors-in-variables Models.

Abstract

Corticomuscular communications are commonly estimated by Granger causality (GC) or directed coherence, with the aim of assessing the linear causal relationship between electroencephalogram (EEG) and electromyogram (EMG) signals. However, conventional GC based on standard linear regression (LR) models may be substantially underestimated in the presence of noise in both EEG and EMG signals: some healthy subjects with good motor skills show no significant GC. In this study, errors-in-variables (EIV) models are investigated for the purpose of estimating underlying linear time-invariant systems in the context of GC. The performance of the proposed method is evaluated using both simulated data and neurophysiological recordings, and compared with conventional GC. It is demonstrated that the inferred EIV-based causality offers an advantage over typical LR-based GC when detecting communication between the cortex and periphery using noisy EMG and EEG signals.