Decoding Forelimb Muscle Activity with Local Field Potentials from Mouse Motor Cortex

Abstract

Electromyography (EMG) decoding is an important tool to study how the cortex controls the muscles of the limbs. Both spike and local field potentials (LFPs) have been used to decode EMG in previous studies where good performances have been achieved in both rats and monkeys. However, it is a big challenge to carry out studies in mice because only a few electrodes are available for neural recording. In this study, we tried to decode the EMG signal from the biceps brachii muscle of the forelimb by using the LFP signals of their motor cortex. When mice were performing the lever-pressing task, the EMG and 4-channel LFP signals were synchronously collected. Three decoding algorithms, Kalman Filter, General Regression Neural Network (GRNN) and Recurrent Neural Network (RNN), were employed to extract the envelope of EMG signals from the LFP signals. Our results showed that all three algorithms are able to achieve good decoding performance even only a few channels were used. In addition, RNN achieved the best decoding performance among these algorithms, whose CC and MSE were 0.83 and 0.013 respectively.