Multilayer Brain Networks for Enhanced Decoding of Natural Hand Movements and Kinematic Parameters

Decoding natural hand movements using Movement-Related Cortical Potentials (MRCPs) features is crucial for the natural control of neuroprosthetics. However, current research has primarily focused on the characteristics of individual channels or on brain networks within a single frequency or time segment, overlooking the potential of brain networks across multiple time-frequency domains. To address this problem, our study investigates the application of multilayer brain networks (MBNs) in decoding natural hand movements and kinematic parameters, using a combination of MRCPs features and MBNs metrics. Based on grasp taxonomy, we selected four natural movements for our study: Large Diameter (LD), Sphere 3-Finger (SF), Precision Disk (PD), and Parallel Extension (PE), each incorporating two levels of speed and force parameters. The results demonstrate that a combination of MRCPs features and MBNs metrics can successfully decode not only the types of movements and kinematic parameters but also differentiate between different grasp taxonomy characteristics, such as the number of fingers exerting force and the type of grasp. In terms of movement type, we achieved a peak four-class accuracy of 60.56%. For grasp type and number of fingers exerting force, binary classification peak accuracies reached 79.25% and 79.28%, respectively. In the case of kinematic parameters, the Precision Disk movement exhibited the highest binary classification peak accuracy at 84.65%. Moreover, our research also found the changes and patterns in brain region connectivity across both time and frequency domains. We believe that our research highlights the potential of MBNs to enhance the functionality of Brain-Computer Interface (BCI) systems for more intuitive control mechanisms and contributes valuable insights into the brain's operational mechanisms.