A Comprehensive Survey on Rehabilitative Applications of Electroencephalogram in Healthcare

Introduction

The brain can be interfaced with a computer to collect brain signals, examine those signals, and convert them into commands given to output devices to perform the expected tasks. The main objective is to restore the useful activity of people who are disabled or less abled by neuromuscular diseases like stroke, spinal cord injury, etc. Brain-computer interfacing (BCI) technology is in focus because it is a quickly growing research and development technology that greatly stimulates scientists, engineers, clinicians, and the public in general. BCI requires convenient, portable, and safe signal acquisition hardware which can work in all situations.

Communication within the human nervous system is done by passing messages between cells using electrochemical signals, resulting in electrical impulses in our brain cells (Ghosh & Saha, 2022). If a subject suffering from prosthesis damage in their arms wants to bring a glass of water from a nearby table to drink, the subject cannot bring it using their arms. A similar problem is faced by subjects with disorders in the brain's parietal lobe when they attempt to perform the same task (Ghosh & Saha, 2020). But in both situations, an electroencephalography (EEG) acquisition device can be used to monitor and record the occipital data of that subject while they are planning to do the activity by looking at the glass, and depending on the recorded EEG signal, a robotic arm (artificial limb) is active to grasp the glass. This simple scenario is controlling an artificial limb based on our brain signals. The work is vital and effective in rehabilitative applications where a subject suffers from parietal and/or motor cortex disorders and, in turn, suffers from prosthesis issues.

Neuronal postsynaptic membrane polarity changes generate various electrical signals due to voltage-gated activation (Nakajima & Baker, 2018). Nowadays, BCI is implemented using non-invasive EEG signals, which are easy, safe, and inexpensive to acquire. These electrical signals are significantly decreased when passing through the dura, skull, and scalp, which is the main disadvantage (Scrivener & Reader, 2022). Because of this, the main information may be dropped.

In recent years, the coordination control scheme of intelligent agents like robots has drawn the attention of many researchers as the work efficiency of robots outperforms the same of human beings in various fields (Corcione et al., 2005). In the real world, robots suffer from uncertainties like external disturbance, parameter perturbations, and unmodelled dynamics, which are being attempted to overcome by different machine learning (ML) tools. Communication with a human-machine interface may be multimodal since the mouse and keyboard are not accepted nowadays as the only input sensory system. Humans should be able to communicate with robots in ways that are as close to the concise, rich, and diverse forms of communication they use with one another as possible.

Several researchers have targeted rehabilitation as the application area of ML. A nonlinear sliding mode controller has been attempted to be implemented for the upper limb exoskeleton in (Rahman et al., 2012) for stroke rehabilitation, where different exercises are to be undergone involving single and multi-joint movements. The experimental outcome has proved it as an effective one. Still, it dealt with only passive arm movements, whereas in the future, active movements are also to be considered to broaden the scope of the control strategy. Respiratory rehabilitation has been the area of research in Lee et al.'s robot assistance work (Lee et al., 2015), where the robot is mainly controlled by an attached joystick which creates pressure in the abdominal region of the human body and helps to exhale easily. Logarithmic profiles with some definite shape parameters have been useful in this case, and user safety has also been another concern. Three degrees of freedom wearable robotic assistive devices have been designed to help the subject in passive rehabilitation (Islam et al., 2017), where movement tracking error is 0.5 degrees.