A novel approach for removing micro-stimulation artifacts and reconstruction of broad-band neuronal signals

Background: Electrical stimulation is a widely used method in the neurosciences with a variety of application fields. However, stimulation frequently induces large and long-lasting artifacts, which superimpose on the actual neuronal signal. Existing methods were developed for analyzing fast events such as spikes, but are not well suited for the restoration of LFP signals. New Method: We developed a method that extracts artifact components while also leaving the LFP components of the neuronal signal intact. We based it on an exponential fit of the average artifact shape, which is subsequently adapted to the individual artifacts amplitude and then subtracted. Importantly, we used for fitting of the individual artifact only a short initial time window, in which the artifact is dominating the superimposition with the neuronal signal. Using this short period ensures that LFP components are not part of the fit, which leaves them unaffected by the subsequent artifact removal. Results: By using the method presented here, we could diminish the substantial distortions of neuronal signals caused by electrical stimulation to levels that were statistically indistinguishable from the original data. Furthermore, the effect of stimulation on the phases of gamma- and beta - oscillations was reduced by 85 and 75 %, respectively. Comparison with Existing Methods: This approach avoids signal loss as caused by methods cutting out artifacts and minimizes the distortion of the signal's temporal structure as compared to other approaches. Conclusion: The method presented here allows for a successful reconstruction of broad-band signals.