EEG-Based Multi-Class Classification for Recognizing Pedaling Velocities: A Promising Approach for Brain-Computer Interface-Enhanced Lower-Limb Robotic Rehabilitation

Lower-limb robotic rehabilitation devices have shown promising results in restoring functional activities of individuals with motor impairments. Motorized Mini-Exercise Bikes (MMEBs) are robotic devices that provide functional and repetitive flexion/extension exercises through Passive Pedaling (PP), improving muscular activity, coordination, and functional independence. These technologies activated by Brain-Computer Interfaces (BCIs) can be useful during therapeutic interventions for individuals who cannot voluntarily perform a movement. This study proposes a multi-class classification strategy for recognizing different pedaling velocities using electroencephalography signals. To provide a meaningful task, we conducted a protocol with ten able-bodied subjects who received PP from an MMEB configured at 30, 45, and 60 rpm. We provide a practical demonstration by implementing a feature extraction stage based on Common Spatial Patterns, Power Spectral Density, and a classification system based on four Machine Learning techniques: Linear Discriminant Analysis, Support Vector Machine, k-nearest Neighbors, and Decision Tree. Our strategy achieved a mean accuracy of 0.86, false positive rate of 0.13, and kappa index of 0.64, we concluded that the system has potential to be used in the design of robust controllers for robotic devices. Further research will be centered on real-time evaluation of our strategy to implement an MMEB-based BCI for personalized rehabilitation of people with lower-limb limitations, such as stroke or spinal cord injury survivors.