Regression-based reconstruction of human grip force trajectories with noninvasive scalp electroencephalography

Abstract

Objective. Robotic devices show promise in restoring motor abilities to individuals with upper limb paresis or amputations. However, these systems are still limited in obtaining reliable signals from the human body to effectively control them. We propose that these robotic devices can be controlled through scalp electroencephalography (EEG), a neuroimaging technique that can capture motor commands through brain rhythms. In this work, we studied if EEG can be used to predict an individual's grip forces produced by the hand. Approach. Brain rhythms and grip forces were recorded from able-bodied human subjects while they performed an isometric force production task and a grasp-and-lift task. Grip force trajectories were reconstructed with a linear model that incorporated delta band (0.1-1 Hz) voltage potentials and spectral power in the theta (4-8 Hz), alpha (8-13 Hz), beta (13-30 Hz), low gamma (30-50 Hz), mid gamma (70-110 Hz), and high gamma (130-200 Hz) bands. Trajectory reconstruction models were trained and tested through 10-fold cross validation. Main results. Modest accuracies were attained in reconstructing grip forces during isometric force production (median r = 0.42), and the grasp-and-lift task (median r = 0.51). Predicted trajectories were also analyzed further to assess the linear models' performance based on task requirements. For the isometric force production task, we found that predicted grip trajectories did not yield static grip forces that were distinguishable in magnitude across three task conditions. For the grasp-and-lift task, we estimate there would be an approximate 25% error in distinguishing when a user wants to hold or release an object. Significance. These findings indicate that EEG, a noninvasive neuroimaging modality, has predictive information in neural features associated with finger force control and can potentially contribute to the development of brain machine interfaces (BMI) for performing activities of daily living.

Deep learning multimodal fNIRS and EEG signals for bimanual grip force decoding

“…We resampled all signals to 80 Hz for the 'raw' EEG stream. Otherwise, in order to extract 'Hilbert features' the EEG was decomposed in the power and phases of the following frequency bands as in [14] We also recorded the scalp skin's hemodynamic activity on the forehead using a NONIN 8000R…”

“…The same decoding approach was used by [14] with some differences. Their approach used the Fourier transform to extract only the power content, not the phases, of each EEG band except for delta.…”

“…We first examine if multimodal signals (fNIRS and EEG) provide an advantage in bimanual continuous force decoding compared to each separate modality. For comparison, we used the linear approach in [14] to do this. Secondly, we explore DL techniques to fuse these multimodal signals with different neurophysiological origins.…”

HYGRIP: Full-Stack Characterization of Neurobehavioral Signals (fNIRS, EEG, EMG, Force, and Breathing) During a Bimanual Grip Force Control Task

“…The lack of mu-ERD and force correlation during the 20 s of continuous contraction/relaxations might be a consequence of the velocity at which the sequential contractions/relaxations were executed, not leaving enough time to the motor cortex to reach a synchronized equilibrium state before it was desynchronized again. Furthermore, this can also indicate that mu frequencies (8-12 Hz) are not fast enough to track this kind of subtle phase changes and a justification to develop more precise algorithms or feature extractors as suggested by Paek et al (2019).…”

“…Non-invasive BCIs have also succeeded in the continuous control of trajectories after users learned to modulate eventrelated desynchronization (ERD) (Wolpaw and McFarland, 2004;Royer et al, 2010;Meng et al, 2016). However, while force is central to motor control (Westling and Johansson, 1984;Ostry and Feldman, 2003), its continuous non-invasive decoding is still challenging even in the offline case, and only modest accuracies have been reported using electroencephalography (EEG) (Paek et al, 2019). Previous attempts at decoding force from non-invasive measures have focused on the classification of discrete force variables using EEG (Jochumsen et al, 2013;Wang et al, 2017).…”