Proprioceptive afferent activities could be useful for providing sensory feedback signals for closed-loop control during functional electrical stimulation (FES). However, most previous studies have used the single-unit activity of individual neurons to extract sensory information from proprioceptive afferents. This study proposes a new decoding method to estimate ankle and knee joint angles using multiunit activity data. Proprioceptive afferent signals were recorded from a dorsal root ganglion with a single-shank microelectrode during passive movements of the ankle and knee joints, and joint angles were measured as kinematic data. The mean absolute value (MAV) was extracted from the multiunit activity data, and a dynamically driven recurrent neural network (DDRNN) was used to estimate ankle and knee joint angles. The multiunit activity-based MAV feature was sufficiently informative to estimate limb states, and the DDRNN showed a better decoding performance than conventional linear estimators. In addition, processing time delay satisfied real-time constraints. These results demonstrated that the proposed method could be applicable for providing real-time sensory feedback signals in closed-loop FES systems.
Bibliographical noteFunding Information:
This work was supported in part by a grant of the Next-generation Medical Device Development Program for Newly-Created Market of the National Research Foundation (NRF) funded by the Korean Government, MSIP (No. 2015M3D5A1066100), the Bio & Medical Technology Development Program of the NRF funded by the Korean government, MSIP (No. 2014M3A9D7070128), and the convergence technology development program for bionic arm through the NRF funded by the Ministry of Science, ICT & Future Planning (No. 2016M3C1B2913050 and 2016M3C1B2913053).
© The Author(s) 2017.
ASJC Scopus subject areas