A novel deep neural network approach incorporating brain-muscle connectivity achieved a prediction accuracy of 96.5% (± 1.2%) on a simulated dataset, surpassing conventional methods.
A novel deep neural network approach explicitly incorporating brain-muscle connectivity improves the prediction accuracy of muscle synergy analysis compared to conventional methods.
Muscle synergy, a complex neuronal connectivity mechanism within the central nervous system encompassing both the brain and muscles, plays a pivotal role in advanced motor control. Its analysis has emerged as vital perspective in assessing rehabilitation progress and complex motor skills. However, traditional methodologies often disregard the explicit utilization of both brain and muscle activities, thus failing to consider the crucial connectivity between them, and frequently treat muscle activity as an isolated entity. We have therefore developed a novel approach to muscle synergy analysis that explicitly incorporates this connectivity. Our method leverages the power of deep neural network models to predict muscle activity from brain activity, thereby capturing the synergy components under the connectivity constraints between them. Our analysis using a simulated dataset demonstrated that the prediction accuracy of the proposed method reached 96.5% (± 1.2%), surpassing other conventional methods. In real-world datasets, the standard deviation of the synergy weights for the proposed method was significantly higher than that observed with Non-negative Matrix Factorization, offering a more precise representation of the underlying neurological connectivity. Our findings can potentially make substantial contributions to the study of complex body movements and the creation of more effective rehabilitation strategies.
Isezaki et al. (Sun,) conducted a other in Motor control and rehabilitation. Deep neural network model incorporating brain-muscle connectivity vs. Conventional methods (e.g., Non-negative Matrix Factorization) was evaluated on Prediction accuracy of muscle activity from brain activity. A novel deep neural network approach incorporating brain-muscle connectivity achieved a prediction accuracy of 96.5% (± 1.2%) on a simulated dataset, surpassing conventional methods.