Neural control strategies for a complex biomechanical system: Primary motor cortex and the hand.

摘要

The hand is a complex biomechanical apparatus with 27 bones, 18 joints, and 39 intrinsic and extrinsic muscles, resulting in over 20 degrees of freedom. Despite significant mechanical coupling between the joints of the hand, humans and non-human primates are able to perform both simple and highly intricate movements of the hand and fingers. Much of this ability is attributed to the neural control mechanisms. Although a large number of cortical and subcortical systems are involved in prehension, the primary motor cortex (M1) plays a critical role in reaching to and grasping an object. The strategy utilized by M1 to control movements of the hand is of considerable interest in the fields of neuroscience and engineering. A major area of debate is whether M1 explicitly controls individual degrees of freedom or more global patterns of movement. To test this hypothesis, two rhesus monkeys were trained to reach and grasp a set of 23 different objects that were designed to systematically vary hand shape. Fourteen joint angles and angular velocities of the hand and fingers were monitored simultaneously with the recording of 81 single cells in the hand area of M1. The joint angles were significantly different across objects during the reach and grasp epochs, indicating that the hand preshaped to match properties of the object to be grasped. There were fewer instances of significant differences in joint angular velocities across objects than for the joint angles, especially during the premove and grasp epochs. Singular value decomposition (SVD) analyses defined a dominant hand shaping pattern that was similar across sessions and monkeys that consisted of simultaneous extension/flexion of the MCP and IP joint angles. The majority of the variation in hand shaping was captured by only a few lower-order eigenvectors (EVs), suggesting that they represent major patterns of hand shaping. In contrast, the higher-order EVs characterize the more detailed movements of the hand because of the smaller amount of variance captured. Linear regression analysis revealed that the firing of many M1 cells (up to 38.6%) was highly correlated with individual joint angles with only limited correlation to joint angular velocities. Typically, a cell's firing was correlated with multiple joints. The firing of M1 cells was also highly correlated to the lower-order temporal weighting vectors (TWs) derived from SVD analyses. Higher-order TWs were not well-represented. In addition, most cells displayed high R2-values for multiple lower-order TWs. Correlations were improved most often by incorporating a temporal lead in the neural firing. This suggests that M1 is involved with the control of dominant hand shaping patterns rather than explicitly controlling details. These findings could be used to develop brain-machine interface algorithms in which signals from M1 are used to control robotic or virtual hands.