Abstract
Objective Extracellular electrical recording of neural activity is an essential tool in neuroscience, and spike sorting is a fundamental step to process the recordings. Thus it requires spike sorting algorithms to perform robustly in the face of noise and perturbation. A few algorithms have been proposed to overcome these challenges and difficulties. However, when noise level or waveform similarity becomes relatively high, their robustness still faces big challenges.
Approach Here, we propose a spike sorting method, using Linear Discriminant Analysis (LDA) for feature extraction and Density Peaks (DP) for clustering. Since DP well adapts to the diverse distribution of spikes, LDA in our method can concentrate on features associated with clustering according to the feedback conveyed by DP and ignores the unassociated features. Through a combination of LDA and DP, our method can maintain a highly robust performance in various complex data situations.
Main results In this study, we compete the proposed density-peaks-based framework with several algorithms. It has demonstrated its high level of sorting accuracy and cluster quality, and outperforms previously established methods in both simulated and real extracellular recordings.
Significance Due to the rapid development of acquisition and recording technology, neuroscience involves more and more complex and precise signal analysis and decoding research, which makes it imperative to adopt one highly robust spike sorting method for preprocessing. Through evaluations, it has been explicitly shown that the proposed algorithm has strong robustness under high noise levels and high similarity of spike waveforms, and can meet research requirements.
Competing Interest Statement
The authors have declared no competing interest.
