Fusion of EEG and EMG signals for classification of unilateral foot movements. (July 2020)

• Record Type: Journal Article
• Authors: Hooda, Neha; Das, Ratan; Kumar, Neelesh
• Is Part Of: Biomedical signal processing and control. Volume 60(2020)
• Publication Date: 2020-07
• Subjects: Electroencephalography (EEG) -- Electromyography (EMG) -- Fusion -- Lower limb -- Wireless foot sensor module; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2020.101990
• Languages: English
• ISSNs: 1746-8094

Abstract

Graphical abstract: Highlights: Fusion of cortical (EEG) and muscular (EMG) signals for task classification. A parallel and cascaded bio-signal based classification method. Demonstrated the classification of four lower limb movements. Presented a novel technique of task onset detection from angle measurements. Abstract: Introduction: The study of motor cortex represents the presence of functional activity around mesial surface during all lower limb movements. Due to this, the problem of classification for intricate lower limb movements is particularly challenging with existing non-invasive technologies, such as electroencephalography (EEG). The other bio-signal used for detection i.e. electromyography (EMG), underlines the factors such as muscular fatigue and spasm as possible hindrance for efficient task based classification. Methods: This work aims to explore the fusion of both, EEG and EMG, sensing modules to identify unilateral lower limb movements. Four channel sets were formed with optimal selection of EEG and EMG channels. The processed bio-signals were analyzed for parallel as well as cascaded classification of five tasks. The performance has been assessed using two parameters, prediction accuracy ( PA ) and computational time ( CT ). Results: The approach successfully classified the five tasks with maximum PA of (96.58 ± 2.37)% and CT of (51.89 ± 1.15)ms for cascaded scheme. The optimal performance has been achieved with PA of (90.06 ± 9.71)% and (89.81 ± 9.41)% for channel-set (Ch) i.e. 7-Ch and 3-Ch, respectively. The resulting CT of (52.82 ± 3.56)ms and (65.38 ± 3.36)ms have been obtained for 7-Ch and 3-Ch, respectively. The parallel scheme resulted in PA of (85.88 ± 3.92)% and (86.16 ± 3.97)% along with CT of (33.23 ± 6.74)ms and (34.80 ± 10.42)ms for 7-Ch and 3-Ch, respectively. Conclusion: The obtained results showed a higher PA for the case of cascaded classification compared to the parallel scheme. Promising results have been achieved, for healthy participants and can be used for future applications of robotic device control and rehabilitation.