Novel method for detecting neural spikes from signals with low signal-to-noise ratio

Jing Wang, Zhou Yan Feng, Xiao Jing Zheng

Research output: Contribution to journalArticlepeer-review


There are usually a large number of small amplitude pulses included in extracellular action potential pulse (i. e. spike) recordings. In order to accurately detect these small spikes from recording signals with low signal-to-noise ratio (SNR) and thereby to increase the number of identified neurons from a single experiment, the present work developed a new spike detection algorithm based on the features of tetrode recording signals. The method firstly extracted the first component of the four channel signals by using principal component analysis (PCA). Then, the nonlinear energy operator (NEO) was applied on the first component to obtain the signals with low noises and enhanced spikes for spike detection by using a threshold method. The detection threshold was determined by a type of two step method to decrease the influences from varied spike firing densities and from large amplitude spikes. The results obtained from both synthetic datasets and experimental recordings demonstrate that the PCA-NEO threshold method can be used to processing signals recorded by microelectrode arrays with tetrode-like high density spacings. It is able to significantly increase the accurate detection ratio of small spikes with low SNR. Especially, the method can identify overlapped spikes effectively. Therefore, the new spike detection method can provide more information for neuronal signal decoding and neural network analysis.

Original languageEnglish (US)
Pages (from-to)941-947
Number of pages7
JournalZhejiang Daxue Xuebao (Gongxue Ban)/Journal of Zhejiang University (Engineering Science)
Issue number5
StatePublished - May 2011
Externally publishedYes


  • Neuronal spike
  • Nonlinear energy operator
  • Principal component analysis
  • Signal-to-noise ratio
  • Tetrode


Dive into the research topics of 'Novel method for detecting neural spikes from signals with low signal-to-noise ratio'. Together they form a unique fingerprint.

Cite this