Neural Signal Compression based on Gaussian Curve Fitting

H.J. Ko, W. Song, J.H. Choi, and T. Kim (Korea)


Neural signal compression, Gaussian curve fitting


As a consequence of increasing neural research activities, neural signals of longer duration and more simulations channels are recorded or transmitted. Recording the amount of the recorded data then becomes increasing. We propose a new compression method, which is based on multi-Gaussian modeling of action potentials, and which preserves the position of local peaks and the shape of spike waveforms in neural signals. First, we detect the spike region using the multi-resolution Teager energy operator (MTEO) method. And then each spike is fitted by a multi-Gaussian curve model, and only the parameters of the model are stored instead of the waveform. The background noise and the spike residual signals are finally compressed into a few of coefficients using auto regressive (AR) modeling. For performance comparison, we calculate using a spike sorting the correct classification ratio (CCR) between the original and the compressed spikes. The results show that the proposed method achieves an average CCR of 0.79 at an average compression ratio (CR) of 45:1. We also show that the proposed method is far more efficient than a conventional method both in CR and CCR.

Important Links:

Go Back