An Unsorted Spike-Based Pattern Recognition Method for Real-Time Continuous Sensory Event Detection from Dorsal Root Ganglion Recording

Abstract

In functional neuromuscular stimulation systems, sensory information-based closed-loop control can be useful for restoring lost function in patients with hemiplegia or quadriplegia. The goal of this study was to detect sensory events from tactile afferent signals continuously in real time using a novel unsorted spike-based pattern recognition method. The tactile afferent signals were recorded with a 16-channel microelectrode in the dorsal root ganglion, and unsorted spike-based feature vectors were extracted as a novel combination of the time and time-frequency domain features. Principal component analysis was used to reduce the dimensionality of the feature vectors, and a multilayer perceptron classifier was used to detect sensory events. The proposed method showed good performance for classification accuracy, and the processing time delay of sensory event detection was less than 200 ms. These results indicated that the proposed method could be applicable for sensory feedback in closed-loop control systems.