The effect of automatic simple thresholding for spike detection from multi-unit recordings on the classification of hitting task in rats

A brain computer interface (BCI) is intended for assisting patients with severe neuromuscular impairments. Human intervention for setting parameters such as detection thresholds during intra-cortical recordings is still unavoidable today. In this study we aimed to investigate the effect of simple automatic thresholding for spike detection from multi-unit recordings on movement classification. Three rats were trained to perform multiple hits on a response paddle. The rats were intra-cortically implanted with 4×4 microwire arrays before data recording was performed. In the beginning of each recording session the noise thresholds were registered. Simple thresholding and automatically calculated thresholds were used for spike detection. A linear discrimination analysis (LDA) a support vector machine (SVM) classifiers were used to classify neural data (`Hit´ or `No-Hit´ classes). We found the automatic threshold had comparable effects to registered thresholds on the classification with no statistical significant difference. We suggest further investigation of the automatic thresholding with continues movement such as decoding the direction and speed of hand movement.