Improving EEG-Based Continuous Grip Force Decoding in Grasp-Lift Tasks by Considering Grip Force Levels

Electroencephalography(EEG)-based brain-computer interfaces(BCIs) for motor control can restore or enhance users’ motor ability. Grasp-lift is one of the critical hand movements in daily life. Continuous decoding of grip force in grasp-lift tasks is significant in developing natural motor control BCIs. According to different grasping target objects, the hand takes different grip force levels in grasp-lift tasks. However, it remains unclear whether the performance of continuous grip force decoding in grasp-lift tasks can be improved if considering grip force levels. To address this issue, we define low, medium and high grip force levels based on object weights in grasp-lift tasks. Then we develop corresponding force-specific decoders named EEGForceNET based on CNN+LSTM to decode continuous grip force using low delta frequency band (0-1 Hz) EEG signals. After that, we conduct pseudo-online experiments on WAY-EEG-GAL dataset to evaluate the decoding performance of force-specific models. Finally, with the improved force-specific models, the average Pearson correlation coefficient(PCC) between reconstructed and recorded grip force trajectories is increased by approximately 0.137. The results suggest that the performance of continuous grip force decoding can be improved by differentiating different grip force levels and that the CNN+LSTM model can be used to decode continuous grip force in motor control BCIs.