EEG Characteristics and Classification Models for Motor Imagery at Different Grip Force Levels
Zhao Jialing
Tianjin Key Laboratory of Information Sensing and Intelligent Control, Tianjin University of Technology and Education, Tianjin, China.
Bian Yan *
Tianjin Key Laboratory of Information Sensing and Intelligent Control, Tianjin University of Technology and Education, Tianjin, China.
*Author to whom correspondence should be addressed.
Abstract
Aims: This study investigated the time-frequency and spatial electroencephalography (EEG) characteristics and classification performance associated with motor imagery at different grip force levels, providing experimental evidence for EEG-based decoding of graded grip-force intentions during hand rehabilitation.
Study design: A within-subject design was adopted. Ten healthy right-handed participants performed right-hand grip motor imagery at 10%, 50%, and 90% of maximum voluntary contraction (MVC).
Place and Duration of Study: The study was conducted at Tianjin University of Technology and Education in August 2025.
Methodology: Event-related spectral perturbation (ERSP) analysis was used to characterize time-frequency and spatial features, and a time-frequency cluster-based permutation test was performed to compare ERSP differences among conditions. Filter bank common spatial pattern (FBCSP) features were used to construct minimum distance to Riemannian mean (MDM), support vector machine (SVM), and MDM-SVM fusion models. Pairwise and three-class classification performance was evaluated using 10 repetitions of 10-fold cross-validation. Accuracy was compared with theoretical chance levels using one-sided Wilcoxon signed-rank tests with Holm correction, and model differences were assessed using the Friedman test. Cohen’s κ was used to evaluate classification agreement beyond chance.
Results: All three grip-force conditions exhibited sensorimotor rhythm modulation, but ERSP differences were not significant after correction for multiple comparisons. MDM and MDM-SVM showed comparable performance, with mean accuracies generally higher than those of SVM. Their mean three-class accuracies were 59.61% and 60.14%, respectively. Only MDM-SVM three-class accuracy was significantly above chance (P=.029). Its Cohen’s κ was 0.403 ± 0.462 and significantly greater than zero (P=.029).
Conclusion: Different grip-force motor imagery conditions showed preliminary evidence of EEG class separability, with MDM and MDM-SVM demonstrating favorable decoding performance. ERSP and classification analyses characterized force-related motor imagery from complementary perspectives.
These findings require validation in larger cohorts.
Keywords: Motor imagery, grip force level, event-related spectral perturbation, Riemannian geometry