Wednesday, July 25, 2018

Detection of Multifiber Neuronal Firings: A Mixture Separation Model Applied to Sympathetic Recordings


By: Can Ozan Tan, J. Andrew Taylor, Albert S. H. Ler, and Michael A. Cohen

IEEE Transactions on Biomedical Engineering, 2009

            Regional blood flow is primarily controlled by sympathetic nervous outflow to the vasculature, and when dysregulation occurs pathological conditions such as sleep apnea, hypertension, heart failure, and metabolic syndrome can arise. Therefore, it is important to cardiovascular research to quantify the sympathetic activity. Within traditional human sympathetic nerve recordings, there are three stages: bandpass filtering, full-wave rectification, and integration with a slow time constant. These are all used to purify the signal and obtain action potentials. Bandpass filtering removes low-frequency content of the signal that is possibly physiologically relevant. Full-wave rectification yields artificial amplification of the filtered recording, and integration averages the sympathetic activity over time. In past years, the signals were read “by hand” and experts had to pick out the specific action potentials themselves. To bypass potential problems with this, wavelet decomposition methods have been designed to denoise sympathetic activity recordings and improve quantification. The primary aim of this study was to derive a generic, fully automated technique for restoring raw, multifiber nerve signals that were buried in high levels of noise and other artifacts.

            To achieve this, researchers proposed an algorithm to minimize signal degradation and it was sensitive to different firing modes of multifiber signals. This algorithm followed the basic steps of 1. Removing line noise and movement artifacts from the recording, 2. Using the characteristic cardiac rhythmicity of sympathetic firing to identify action potential candidates and create a mean action potential template (helped identify noise spikes) and 3. Identify actual spikes and background noise. To validate the technique, the researchers produced 5 minute long 10-KHz artificial datasets with various signal-to-noise ratios (SNR). The two types of common artifacts within the raw multifiber recordings were 60-Hz line noise and movement artifacts. The 60-Hz line noise was removed by notch filtering and the movement artifacts were removed by clipping cutoff points in certain quartiles of the normal distribution to fit the time domain. In addition, they considered that bursts within the recording that exceeded 2.5 standard deviations of the whole neurogram were most likely the ones that contained the action potentials.

            The results of the study showed that the new technique produced performance improvement that was statistically significant. To further test the algorithm, they applied to it to renal sympathetic nerve recordings from freely moving rats under baseline conditions. High SNR ensured that there was successful detection of nerve activity, despite the fact that they had a low sampling rate. This showed that they could use the algorithm in species other than humans, and also apply it to multifiber recordings from different types of nerves.

            The researchers successfully described a new algorithm that is capable of restoring multifiber nerve signals buried within levels of background noise. They did consider a few limitations: 1. When the rate goes above 400 spikes per second, their methods tended to overestimate the number of spikes and 2. The method requires a relatively high sampling rate for recordings. Overall, the artificial datasets produced results that showed the algorithm providing an accurate measure of nerve signal, and it actually provided better quantification of the sympathetic signal compared to traditional methods. The major improvements with the new algorithm include that it is able to minimize signal degradation and retain more information about nerve firing.

This is important to our lab because the algorithm is capable of detecting nerve signals in recordings that are heavily contaminated with background noise. Sending our SSNA recordings to this lab will allow them to help us determine if glutamate activation of the RVLM causes more neurons to fire within the nerve, or if it causes the same neurons to fire faster.

-L. Matus

No comments:

Post a Comment