چكيده به لاتين
Providing robust and natural command signals together with the feedback information is a pivotal issue to develop neuroprosthetic devices in clinical platform. The main objective of this research is to record both sensory and motor information from the ascending and descending tracts within the spinal cord for decoding the hindlimb kinematics during walking. To investigate the distribution of kinematic information within gray matters of different segments of the spinal cord, we conducted acute experiments. During these experiments, three single-electrodes were positioned in the right dorsal horn of the spinal cord of the eight cats at L4, L5, and L6 vertebrae while three ipsilateral hindlimb joint angles (hip, knee, and ankle) were passively flexed and extended. We recorded both neural and kinematic signals simultaneously and analyzed them. The results showed that for the hip, knee, and ankle movements, the most information and the best decoding performance is achieved by the L4, L5, and L6 vertebral segments, respectively. For decoding purpose, we proposed a probabilistic model based on recursive Bayesian equations that can be implemented by using a recurrent neural network (RNN) structure. We evaluated the decoding capability of the proposed model by comparing it with conventional RNN and Kalman filter. The decoding performance on experimental data showed that the proposed PRNN improved R2 values significantly (p < 0.05) by 0.03 and 0.11 with respect to RNN and Kalman filter, respectively.
To extract the hindlimb kinematic information from motor and sensory tracts within the spinal cord, we performed chronic recording experiments. To this end, we designed and fabricated four and eight channel arrays and implanted them in the dorsal column (DC) and lateral column (LC) of the L4 segment on the left side. Two different experimental paradigms (i.e., active and passive) were performed in these experiments on the five adult male cats on different days. During active experiments, cats were trained to walk freely on the treadmill, while during passive experiments the limb was passively moved by the experimenter. Local field potential (LFP) activity was recorded and elastic-net regression was employed to estimate the hindlimb kinematics. The results showed that during walking on the treadmill, the hip, knee, and ankle joint angles could be decoded from signals recorded from both sensory and motor tracts with averaged R2 values of 0.52, 0.33, and 0.34, respectively. We also performed information analysis which showed that during the active experiment there is no significant difference between signals recorded from DC and LC, but the information content of the DC is significantly higher than that of the LC during passively applied movement of the hindlimb in the anesthetized cats. This difference could be considered as evidence of the existence of motor information in the signals recorded during awake walking on the treadmill. On two different cats we recorded neural signals from both left and right sides of the spinal cord. Results showed that the decoding performance obtained from both sides are almost equal. Based on analysis on modular and hierarchical models, we showed that kinematic synergy could be decoded from spinal signals. Also, we showed that the modular models with two-level learning approaches could be used for information combination and also for improving the decoding accuracy.
