The invention discloses a neural pulse rapid sorting method and
system based on a multi-stage self-adaptive strategy. The neural pulse rapid sorting method and
system are suitable for real-
time processing of MEA, multi-channel
electrophysiology, deep brain recording and brain-computer interface data. The
system comprises a data import module, a preprocessing module, an adaptive threshold detection module, a waveform
feature extraction module, a multi-scale clustering module, a template updating module, a quality evaluation module and a result export module. According to the system, a dual-channel parallel high-pass filtering and self-adaptive MAD threshold value method is adopted, and real-time pulse detection which is free of artificial participation and can be directly carried out through CPU peak-splitting sorting is achieved; in the clustering stage, UMAP dimension reduction and HDBSCAN dense clustering are combined, and
noise drift self-calibration in continuous records is supported; the template updating strategy is based on
dynamic time warping (DTW) and Bayesian fusion, and can automatically eliminate false peaks, merge similar templates and split
aliasing units. The system supports large-scale
batch processing accelerated by the GPU, and the
data processing time of 32 channels and the sampling rate of 20 kHz in one hour is shortened to 1-2 minutes. The method solves the problems that the speed is low, parameters highly depend on expert experience, drift correction is insufficient and
repeatability is poor in existing spike sorting, and can be applied to
neuroscience research, brain-computer interface and
disease mechanism research and clinical
electrophysiology monitoring.