An Efficient Adaptive Online Neural Spikes Detection and Classification Engine Based on Bayesian Inference

摘要

A new method, called Bayesian inference-based template matching (BIBTM) method, is proposed in this article, which is designed to detect and classify neural spikes from real neural signals. Through this spike detection and classification method, the templates do not need be given in advance, and they can be automatically generated. To evaluate the performance of our method, we built signals with diverse signal-to-noise ratios and firing rates, and also researched two spike template generation methods. Based on the experimental results and comparison, BIBTM method has excellent detection performance. The true positive rates (TPR) and false positive rates (FPR) of the spike detection can reach 0.92 and 0.05 respectively, and the average FPR and average TPR of the spike classification can reach 0.05 and 0.6 respectively. From the discussion and analysis, our proposed BIBTM method not only has high detection and classification accuracy, but also has a simple structure and low complexity