Machine Learning Evaluation of Passive Wireless Neurosensing Recorder for Biopotentials Recognition

Neuropotentials monitoring can help individuals to significantly enhance their physical and mental well-being. We present an evaluation of a multichannel, passive and fully implantable wireless neurosensing system (WiNS). WiNS employs radiofrequency and optical communications to address the need for non-battery-operated systems. In this study, we will present a new automated technique to identify signal segments eliminating the difficulty of manual classification of evoked biopotentials. In addition, machine learning algorithms are adopted to evaluate signal quality from WiNS and compare it with a commercially available wired system. Somatosensory evoked potential data measured from wired and our wireless systems shows < 6% deviation in machine learning testing accuracy, indicating successful detection of biopotential signal as low as μV
pp
. These results support the concept that real-time machine interface for wireless and passive acquisition of biopotentials is indeed feasible translating to several uses for future clinical research.