Two-stage regression of high-density scalp electroencephalograms visualizes force regulation signaling during muscle contraction. (21st August 2019)

• Record Type: Journal Article
• Authors: Hayashi, Masaaki; Tsuchimoto, Shohei; Mizuguchi, Nobuaki; Miyatake, Mizuki; Kasuga, Shoko; Ushiba, Junichi
• Is Part Of: Journal of neural engineering. Volume 16:Number 5(2019:Oct.)
• Publication Date: 2019-08-21
• Subjects: electroencephalogram -- brain-computer interface -- principal component analysis -- multiple linear regression -- force-matching task; Neurosciences -- Periodicals; Biomedical engineering -- Periodicals; 612.8
• DOI: 10.1088/1741-2552/ab221a
• Languages: English
• ISSNs: 1741-2560

Abstract

Abstract: Objective . A critical feature for the maintenance of precise skeletal muscle force production by the human brain is its ability to configure motor function activity dynamically and adaptively in response to visual and somatosensory information. Existing studies have concluded that not only the sensorimotor area but also distributed cortical areas act cooperatively in the generation of motor commands for voluntary force production to the desired level. However, less attention has been paid to such physiological mechanisms in conventional brain-computer interface (BCI) design and implementation. We proposed a new, physiologically inspired two-stage decoding method to see its contribution on accuracy improvement of BCI. Approach . We performed whole-head high-density scalp electroencephalographic (EEG) recording during a right finger force-matching task at three strength levels (20%, 40%, and 60% maximal voluntary contraction following a resting state). A two-stage regression approach was employed that decodes muscle contraction level from EEG signals in the multi-level force-matching task and translates them into: (1) presence/absence of muscle contraction as a first stage; and (2) muscle contraction level as a second stage. Dimensionality reduction of the EEG signals, using principal component analysis, avoided multicollinearity during multiple regression, and data-driven stepwise multiple regression identified EEG components that were involved in the multi-level force-matching task. Main results . An alternatively tuned two-stage regressor accurately decoded muscle contraction level with online processing rather than the conventional decoders, and identified EEG components that were related to voluntary force production. Relaxation/contraction state-dependent EEG components were localized dominantly in the contralateral parieto-temporal regions, whereas multi-level force regulation-dependent EEG components came from the fronto-parietal regions. Significance . Our findings identify respective cortical signalings during relaxation/contraction and multi-level force regulation using a sensor-based approach with EEG. Simulation-based assessment of the current physiologically inspired decoding technique proved improved accuracy in online BCI control.