Abstract—Information in the nervous system is coded as firing patterns of electrical signals called action potential or spike so an essential step in analysis of neural mechanism is detection of action potentials embedded in the neural data. There are several methods proposed in the literature for such a purpose. In this paper a novel method based on empirical mode decomposition (EMD) has been developed. EMD is a decomposition method that extracts oscillations with different frequency range in a waveform. The method is adaptive and no a-priori knowledge about data or parameter adjusting is needed in it. The results for simulated data indicate that proposed method is comparable with wavelet based methods for spike detection. For neural signals with signal-to-noise ratio near 3 proposed methods is capable to detect more than 95% of action potentials accurately. Keywords—EMD, neural data processing, spike detection, wavelet decomposition. I. INTRODUCTION NFORMATION in the nervous system is coded as firing patterns of action potentials, so action potential detection from neural data is essential in the interpretation of neural mechanisms. Neural data composed of spikes and background noise which the later is a combination of unwanted signals due to fluctuations of energy carriers like ions or electrons and action potentials produced by neurons in far field. Because background noise consist of action potentials so spectral analysis methods based on Fourier transform aren't efficient in neural data analysis [1]. So far several methods have been developed for detecting spikes embedded in neural data. The simplest and most convenient method is spike detection based on simple thresholding. In the case of low signal to noise ratios (SNRs) the efficiency of such method is significantly poor. Also recognition of overlapped spikes is impossible in simple thresholding method [2]. Methods based on neural network have been utilized for spike detection [3] but neural networks need a-priori information about signal and noise characteristics for training purposes which aren't always available in neural data processing, especially in low SNR cases. Another technique for spike detection is template matching which detects spikes based on the similarity between S. Farashi is Phd student in Shahid Beheshti University of Medical Sciences, Faculty of Medecine, Tehran, Iran (e-mail: farashi@sbmu.ac.ir). MD. Abolhassani is with Medical Physics and Biomedical Engineering Department, Tehran University of Medical Sciences, Tehran, Iran (e-mail: abolhasm@sina.tums.ac.ir). M. Taghavi Kani was with Medical Physics and Biomedical Engineering Department, Tehran University of Medical Sciences, Tehran, Iran (e-mail: mtaghavi@razi.tums.ac.ir). neural signal and a predefined template. The result of such method outperforms than simple thresholding but its performance highly depends on the template selection and predefined threshold for similarity measurements [4]. Transforms like wavelet are other choices [2]-[5] to map neural data to transform space and search the presence of action potentials in that space. However in wavelet domain select a suitable wavelet is always a question and must be survived. For example comparison between discrete wavelet transforms (DWT) and stationary wavelet transform (SWT) indicates that SWT outperforms than DWT in spike detection [5]. In multiresolution wavelet domain the method based on multiplication of wavelet coefficients in some successive detail levels has been proposed in [1] which relies on the band limited properties of action potential. This method is sensitive to a threshold for decision. For solving spike detection, method based on high order statistics has been proposed in [6]. Based on this assumption, the background noise is Gaussian in nature [7], statistics with order higher than two can be used to separate Gaussian noise and spikes. In this paper an adaptive method based on EMD has been developed for action potential detection in an automated manner. The method is adaptive and needs no a-priori information about neural data. We have shown it is comparable with wavelet based methods. II. MATERIALS AND METHOD A. Recording from Cockroach A single tungsten microelectrode with impedance about 1MΩ, inserted in the cockroach body, has been employed for recording real neural data. During recording, cockroach has been restrained firmly on a plastic disk. Cockroach only enables to move its antenna freely. This causes fewer artifacts to be induced on a recorded signal. The plastic disk is located on a faraday cage for electromagnetic interference reduction. The signal is applied to an electrophysiological amplifier set through a preamplifier, consist of TLC2272AC (Texas instruments, USA). Analog neural data has been amplified with a gain equal to 2000 and filtered in the range of 0.3-3 kHz. A data acquisition card is utilized to digitize analog amplified and filtered data. The sampling frequency is adjusted to 30ksample/s for satisfying nyquist theorem. Using NI-Labview8.6 (National instruments, USA) software has been prepared for controlling data acquisition includes saving and displaying acquired data. This software controls sampling frequency of data acquisition card. To increase data sample Sajjad Farashi, Mohammadjavad Abolhassani, Mostafa Taghavi Kani I An Empirical Mode Decomposition Based Method for Action Potential Detection in Neural Raw Data World Academy of Science, Engineering and Technology International Journal of Medical and Health Sciences Vol:8, No:1, 2014 45 International Scholarly and Scientific Research & Innovation 8(1) 2014 ISNI:0000000091950263 Open Science Index, Medical and Health Sciences Vol:8, No:1, 2014 publications.waset.org/9997436/pdf