Analysis of EEG signals using deep learning to highlight effects of vibration-based therapy on brain

Abstract

Rehabilitation of patients with neurological disorders is a life-long process accomplished through pharmacological and non-pharmacological procedures for patient management. One non-pharmacological treatment that has become increasingly popular in the last decade is brain stimulation using vibrations. This study analyzes the impact of vibrations on the brain utilizing electroencephalogram (EEG) signal analysis and visualizations to determine how they affect the brain. The participants were first treated with a controlled procedure by giving an illusion of vibrations, and then these subjects had an intervention phase during which vibrations were administered. EEG signals of the individuals were recorded in both phases during the pre-post therapy period using a 32-channel cap following a defined protocol. In contrast to the control group, topographical maps of five frequency bands of the intervention group showed activation in frontal theta and contra-lateral beta activities with further evidence provided by the outcomes of a paired sample t-test and Boxplots. The sensitivity of topo maps cannot reflect minor changes in the visualizations, although there is activity in the brain at that time. Therefore, the classification of EEG signals into controlled and intervention groups is performed to highlight the significance of those minor changes in EEG signals. This study proposed 3D Convolutional Neural Network (CNN) architecture with a combination of efficientnet-b4 for the classification of minor changes among these groups based on five frequency bands separately and combined clean data. The highest accuracy of 100% is achieved with combined clean data, beta, delta, theta, and gamma bands, while 98.34% for the alpha band.