۲- Novel signal processing methods for few EEG electrode-based neural decoding

۳- Remaining challenges and future directions

References

Introduction

The brain is the commander of voluntary movement control. It generates the oscillatory neural activity at specific frequency bands traveling through the corticospinal tract to activate the musculoskeletal system for movement execution [1]. Neural oscillations at mu (8e13 Hz) and beta (15e35 Hz) bands measured around the sensorimotor cortex, known as cortical sensoryemotor rhythms (CSMRs), are thought to be associated with voluntary control of movements [2]. The coupling between CSMR and muscle activities has been previously reported at these frequency bands, confirming the key functional role of CSMR in movement control [3,4]. Damage to the corticospinal tract following a brain or spinal injury can result in a decrease in the coupling between CSMR and muscle activities, and associated motor impairments, such as muscle weakness and loss of independent movement control [5,6]. However, the CSMR may be preserved at the sensorimotor cortex that allows the identification of motor intentions via measuring and decoding the CSMR [7].

Electroencephalography (EEG) is an electrophysiological monitoring technique that records the oscillatory cortical activity including the CSMR. By placing the electrodes on the scalp, the EEG measures cortical activity without surgery. Compared to other brain signal recording methods (e.g., functional MRI, electrocorticography, positron emission tomography), the advantages of EEG are that it is inexpensive, low-risk, and portable [8]. These advantages allow EEG to be an online monitoring method for daily use. However, due to the volume conduction through the scalp, skull, and other layers of the brain, EEGs recorded by a scalp sensor are a “blurred” copy of multisource activities, which increases the difficulty of EEG signal decoding. Advanced signal processing methods are required to address this challenge. Traditional signal processing methods such as independent component analysis [9] and common spatial pattern (CSP) filter [10,11] need a large number of EEG electrodes covering the whole scalp during the measurement for disentangling the mixed multisource signals. This whole-scalp recording reduces the feasibility of EEG in neuro-rehabilitation for daily use. Several novel signal processing methods have been recently proposed to improve EEG data analysis for accurate identification of motor intentions using only a few electrodes. These novel methods can be combined with various new EEG devices with very few electrodes, such as the Emotiv Epoc headset and LooxidVR package for daily use.