Abstract
<title>Abstract</title> <p>Existing motor imagery brain–computer interfaces mainly discriminate coarse-grained movement categories, such as left-hand, right-hand, and foot movements, and therefore have limited ability to represent graded grip-force intentions during hand rehabilitation. To characterize the neural representations and decodability of motor imagery at different grip-force levels, 12 healthy participants performed a motor execution–guided motor imagery task under 10%, 50%, and 90% of maximum voluntary contraction. Movement-related cortical potentials, event-related spectral perturbations, and the directed phase lag index were used to examine time-domain, time–frequency, and directional functional connectivity features across force conditions. The classification performance of EEGSym was compared with that of FBCSP-SVM, FBCSP-MDM, and a probability-weighted fusion model. The results showed a significant condition effect in the movement-related cortical potential at C4, more pronounced beta-band event-related desynchronization over the central sensorimotor region under the high-force condition, and significant differences in directional functional networks in the delta, alpha, and gamma bands. EEGSym achieved the highest classification accuracy among the compared models. Its accuracies for low-versus-high and medium-versus-high force classification were 77.81% ± 18.90% and 72.60% ± 15.23%, respectively, while the three-class accuracy reached 62.01% ± 21.82%. All of these tasks performed significantly above their corresponding theoretical chance levels. These findings indicate that motor imagery at different force levels produces distinguishable neural representations across slow cortical potentials, sensorimotor rhythms, and directional functional networks. EEGSym can effectively capture spatiotemporal features associated with subtle force differences, providing a basis for decoding force-graded motor intentions in rehabilitation robots, intelligent prostheses, and biomimetic robotic hands.</p>