Decoding hind limb kinematics from neuronal activity of the dorsal horn neurons using multiple level learning algorithm

Hamed Yeganegi, Yaser Fathi, Abbas Erfanian
2018 Scientific Reports  
Decoding continuous hind limb joint angles from sensory recordings of neural system provides a feedback for closed-loop control of hind limb movement using functional electrical stimulation. So far, many attempts have been done to extract sensory information from dorsal root ganglia and sensory nerves. In this work, we examine decoding joint angles trajectories from the single-electrode extracellular recording of dorsal horn gray matter of the spinal cord during passive limb movement in
more » ... ized cats. In this study, a processing framework based on ensemble learning approach is propose to combine firing rate (FR) and interspike interval (ISI) information of the neuronal activity. For this purpose, a stacked generalization approach based on recurrent neural network is proposed to enhance decoding accuracy of the movement kinematics. The results show that the high precision neural decoding of limb movement can be achieved even with a single electrode implanted in the spinal cord gray matter. Restoration of paralyzed extremities through functional electrical stimulation (FES) is a presented paradigm for individuals with neuromuscular disorders and spinal cord lesions. In FES methodology, by applying electrical stimulations to cause contractions in the paralyzed muscles and restoring some mobility, not only an improvement in the independence of disabled people is attained but also their general health conditions are ameliorated 1 . In order to take advantage of the benefits of a closed loop scheme in controlling complex limb movements, a continuous feedback of limb states is needed. Cutaneous and proprioceptive afferents have been proposed as a natural source of sensory feedback for FES systems 2 . Stein et al. investigated the possibility of extracting the position of the foot in space (positions and velocities in Cartesian (x, y) and polar coordinates) from populations of neurons in the dorsal root ganglion (DRG) 3 . For this purpose, they recorded neural signals from up to 100 discriminable nerve cells in the L6 and L7 DRG of the anesthetized cat and employed a linear filter to decode the end-point of the limb in space from the firing rates (FRs) of the sorted neurons. It was reported that predictions using only the one neuron, whose firing was best correlated to the kinematic variable, accounted for about 70% of the variance and it was demonstrated that as more neurons were added, the performance increased and reached a plateau. To find the most informative neurons from a large population of identified neurons, a heavy offline processing is required. This makes the approach difficult for real-time control applications. Moreover, the decoding is based on a linear model. However, the firing rates of individual neurons do not necessarily need to be linearly related to the kinematics 4,5 . Decoding the hind limb kinematics (i.e., ankle, knee, and hip joint angles) from the neural activity of a few neurons in the L7 dorsal root ganglia of three cats has been investigated during walking using a linear filter 2,6 . To improve the decoding performance, a nonlinear state space model was also employed for decoding the FRs in an ensemble of populations of primary afferent neurons during passive movements 4 . In 5 a neuro-fuzzy neural network (FNN) was applied for decoding DRG recordings to estimate limb kinematics during passive as well as voluntary limb movements in cats. It was demonstrated that FNN model provided more accurate estimates of limb state and generalized better than multiple linear regression methods. Reconstruction of forelimb kinematic variables from neural activity of DRG neurons has been also studied in monkeys during voluntary reach-to-grasp
doi:10.1038/s41598-017-18971-x pmid:29330489 pmcid:PMC5766487 fatcat:jzvkxjb4xje7baxp7fjceabpze