DETECTION AND SORTING OF ACTION POTENTIALS RECORDED BY MULTI-ELECTRODE SYSTEM
To understand the population behaviors in nervous system, we need both the real-time information each neuron carries, and the exact firing sequence of the individual neuron. Multi-electrode system is the fundamental tool for research in nervous system. With its development after years, multi-electrode system has been widely used. But the process of the signal extracted from multi-electrode system is still a big problem. We want the exact firing sequence of individual neuron, however, the signal is corrupted with a large amount of background noise and the signal may involve the firing activities of more than one neuron. To get all the spikes, count the number of neurons contributing to the signal, and find the neuron that fires each spike is the very first step of all the research. Many methods have been developed, but the first step, spike detection is still not satisfying. Therefore, in this paper, a better method of spike detection is expected.
This paper proposes a method combining threshold detection and morphological filter. Firstly, apply the threshold detection to the recorded data, and cluster the spikes with principal component analysis. Secondly, take the template of one cluster of spikes as the morphological filter. Finally, filter the raw data with the best morphological filter and redetect the filtered signal. This method works better than threshold detection with varied signal noise ratio. When the baseline shifts, the new method is not influenced while the threshold detection is apparently inferior. The method is also used to process real data.