Predicting the tasks of disabled persons using deep learning-based motor imagery model in BCI applications

• Published in: Research on biomedical engineering 2023-12, Vol.39 (4), p.977-989
• Main Authors: Nayak, Pinki, Meenakshi, S., Medikondu, Nageswara Rao
• Format: Article
• Language: English
• Subjects: Biomaterials; Biomedical Engineering and Bioengineering; Biomedical Engineering/Biotechnology; Engineering; Original Article
• ISSN: 2446-4740; 2446-4740
• DOI: 10.1007/s42600-023-00321-8

Purpose In order to support the disabled persons, researchers developed a motor imagery concept as an important tool. By using this concept, the disabled persons can do daily activities with the help of brain-computer interface (BCI) applications. However, the motor imagery movements of simple limb discrimination in BCI is a challenging concept, even though various machine learning (ML)/deep learning (DL) techniques are developed. The MI model plays a crucial role in BCI systems designed for various applications, such as assistive technologies, neurorehabilitation, and controlling external devices. Here are some key purposes of the MI model in BCI applications. Methods According to the DL concepts, this research work constructs a system model based on motor imagery electroencephalogram (EEG) signals. Two datasets such as BCI-Competition-III-IVa (Dataset A) and BCI-Competition-IV-1 (Dataset B) are considered as input for proposed model, where pre-processing plays a major role in this work. Conclusion In conclusion, the motor imagery (MI) model is a fundamental component of brain-computer interface (BCI) applications that harnesses the power of the brain’s motor imagery to enable communication, control, and interaction for individuals with diverse abilities. By interpreting the brain signals associated with imagined movements, the MI model transforms mental intentions into actionable commands, facilitating a range of practical and impactful applications The Common Average Reference (CAR) filter and Laplace spatial filtering (LSF) are used as pre-processing model to remove noise and filters the components of frequency, which are not considered in the motor imagery function. Result Then, this pre-processed signal is given as an input to the new pre-trained model of convolutional neural network (CNN) for prediction tasks. The experimental analysis provided that the proposed model of CNN achieved 83.83% of average accuracy on Dataset A and 83.88% of average accuracy on Dataset B.