Object-dependent Sparse Representation for Extracellular Spike Detection.

Guangming Shi,Zuozhi Liu,Xiaotian Wang,Chengyu T. Li,Xiaowei Gu

DOI: https://doi.org/10.1016/j.neucom.2017.05.082

IF: 6

2017-01-01

Neurocomputing

Abstract:Extracellular neural recordings have become an essential technique for studying and monitoring neural activity, which is of vital importance in understanding brain functions. However, how to accurately and effectively detect spikes from the extracellular recording signals is still a major related challenge. The existing methods for spike detection have achieved much progress while they are vulnerable to the background noise. In this paper, we develop an object-dependent sparse representation framework for high accuracy and robust extracellular spike detection. Specifically, by exploiting the structural similarities of spikes, we construct an object-dependent dictionary to achieve a sparse and comprehensive representation of the recorded signals. Thus, the problem of spike detection can be formulated as a convex sparse optimization problem. Through systematically analyzing the optimal solution, the number and locations of spikes in the recorded signal are finally determined. In addition, singular value decomposition (SVD) is introduced to further improve the flexibility and robustness of the proposed method. Experimental results on both synthesized extracellular neural recordings and real data show that the proposed method outperforms the existing methods.