The Effects of Filter's Class, Cutoff Frequencies, and Independent Component Analysis on the Amplitude of Somatosensory Evoked Potentials Recorded from Healthy Volunteers

Abstract

Objective: The aim of this study was to investigate the effects of different preprocessing parameters on the amplitude of median nerve somatosensory evoked potentials (SEPs).

Methods: Different combinations of two classes of filters (Finite Impulse Response (FIR) and Infinite Impulse Response (IIR)), three cutoff frequency bands (0.5-1000 Hz, 3-1000 Hz, and 30-1000 Hz), and independent component analysis (ICA) were used to preprocess SEPs recorded from 17 healthy volunteers who participated in two sessions of 1000 stimulations of the right median nerve. N30 amplitude was calculated from frontally placed electrode (F3).

Results: The epochs classified as artifacts from SEPs filtered with FIR compared to those filtered with IIR were 1% more using automatic and 140% more using semi-automatic methods (both p < 0.001). There were no differences in N30 amplitudes between FIR and IIR filtered SEPs. The N30 amplitude was significantly lower for SEPs filtered with 30-1000 Hz compared to the bandpass frequencies 0.5-1000 Hz and 3-1000 Hz. The N30 amplitude was significantly reduced when SEPs were cleaned with ICA compared to the SEPs from which non-brain components were not removed using ICA.

Conclusion: This study suggests that the preprocessing of SEPs should be done carefully and the neuroscience community should come to a consensus regarding SEP preprocessing guidelines, as the preprocessing parameters can affect the outcomes that may influence the interpretations of results, replicability, and comparison of different studies.

Keywords: EEG; ICA; SEPs; filtering; preprocessing.

Figure 1

Methodology overview. The colors group similar processes or sub-processes. Blue is filter properties, yellow corresponds to artifact detection and rejection, salmon pink represents steps related to independent component analysis (ICA), dark green are cleaned datasets, and red are related to somatosensory evoked potential (SEP) averaging and amplitude. Abbreviations: FIR = Finite Impulse Response; IIR = Infinite Impulse Response; AMICA = adaptive mixture ICA; IC = Independent Component.

Figure 2

N30 amplitude. Dots represent N30 amplitude of each dataset. Boxplots show the median, 25th and 75th percentiles. The error bars represent mean ± 95% CI. The distribution plots show the density distribution estimated by a Gaussian kernel with SD of 1.5. The figure was made using the code provided by [29].

Figure 3

SEPs. Grand average SEPs filtered with (A) FIR and (B) IIR. Mean SEPs from one session of a representative participant processed with (C) FIR and (D) IIR filter.

Figure 4

The effect of filter class. The error bar shows estimated mean N30 amplitude ± 95% CI. The class of filter (FIR or IIR) had no effect on the N30 amplitude.

Figure 5

The effect of cutoff frequency and the use of ICA. The error bar shows estimated mean N30 amplitude ± 95% CI. The 30–1000 Hz band showed significantly lower N30 amplitude compared to the 0.5–1000 Hz and 3–1000 Hz bands. The use of ICA significantly reduced the N30 amplitude.