Reliability of motor and sensory neural decoding by threshold crossings for intracortical brain–machine interface

作者：Jun Dai, Peng Zhang, Hongji Sun, Xin Qiao, Yuwei Zhao, Jinxu Ma, Shaohua Li, Jin Zhou, Changyong Wang · 发表于：Journal of Neural Engineering · 年份：2019 · DOI：10.1088/1741-2552/ab0bfb · 被引用次数：37 · 研究领域：EEG and Brain-Computer Interfaces、Neural dynamics and brain function、Functional Brain Connectivity Studies

OBJECTIVE: For intracortical neurophysiological studies, spike sorting is an important procedure to isolate single units for analyzing specific functions. However, whether spike sorting is necessary or not for neural decoding applications is controversial. Several studies showed that using threshold crossings (TC) instead of spike sorting could also achieve a similar satisfactory performance. However, such studies were limited in similar behavioral tasks, and the neural signal source mainly focused on the motor-related cortical regions. It is not certain if this conclusion is applicable to other situations. Therefore, we compared the performance of TC and spike sorting in neural decoding with more comprehensive paradigms and parameters. APPROACH: Two rhesus macaques implanted with Utah or floating microelectrode arrays (FMAs) in motor or sensory-related cortical regions were trained to perform a motor or a sensory task. Data from each monkey were preprocessed with three different schemes: TC, automatic sorting (AS), and manual sorting (MS). A support vector machine was used as the decoder, and the decoding accuracy was used for evaluating the performance of three preprocessing methods. Different neural signal sources, different decoders, and related parameters and decoding stability were further tested to systematically compare three preprocessing methods. MAIN RESULTS: TC could achieve a similar (-4.5 RMS threshold) or better (-3.0 RMS threshold) decoding performance compare...