Surrogate deep neural networks reveal hierarchical handwriting encoding in the human motor cortex

Skilled fine movements are essential for daily life. Although prior work has identified motor cortical tuning to low-level kinematic features like velocity and position, these findings fall short of explaining the precision underlying complex motor behaviors. Critically, it remains unclear whether and how the motor cortex (MC) represents higher-level features of movement. Using single-unit recordings from the human MC during handwriting, we employed surrogate deep neural networks (DNNs) as a tool to investigate these mechanisms. We found that surrogate DNNs capture key aspects of neural activity at both single-unit and population levels. Through this approach, we demonstrate that the MC encodes hierarchical information of movement, including both low-level kinematics and high-level features related to the written content. These results uncover neural encoding behind dexterous motor execution and provide a framework for studying the neural basis of complex behavior.