Simulated Eyeblink Artifact Removal with ICA: Effect of Measurement Uncertainty

TL;DR

The paper addresses how measurement uncertainty in EEG hardware affects ICA-based eyeblink artifact removal by first validating that ADC-related uncertainty is Gaussian and then evaluating FastICA and Infomax across multiple electrode configurations using synthetic eyeblink injections. Using Monte Carlo simulations with additive Gaussian noise across $SNR$ levels, it shows that both algorithms produce similar eyeblink-identification performance, with degradation remaining below about $5\%$ for $SNR$ above $15$ dB. A key practical finding is that FastICA’s execution time decreases with higher uncertainty (50–85% faster across a $20$ dB range) due to its sequential component extraction, while Infomax remains largely unaffected by measurement uncertainty and depends mainly on channel count. The results inform guidelines for selecting ICA algorithms and electrode sets in real EEG/BCI artifact removal tasks, balancing accuracy and computational efficiency under realistic measurement conditions.

Abstract

Independent Component Analysis (ICA) is commonly-used in electroencephalogram (EEG) signal processing to remove non-cerebral artifacts from cerebral data. Despite the ubiquity of ICA, the effect of measurement uncertainty on the artifact removal process has not been thoroughly investigated. We first characterize the measurement uncertainty distribution of a common ADC and show that it quantitatively conforms to a Gaussian distribution. We then evaluate the effect of measurement uncertainty on the artifact identification process through several computer simulations. These computer simulations evaluate the performance of two different ICA algorithms, FastICA and Infomax, in removing eyeblink artifacts from five different electrode configurations with varying levels of measurement uncertainty. FastICA and Infomax show similar performance in identifying the eyeblink artifacts for a given uncertainty level and electrode configuration. We quantify the correlation performance degradation with respect to SNR and show that in general, an SNR of greater than 15 dB results in less than a 5% degradation in performance. The biggest difference in performance between the two algorithms is in their execution time. FastICA's execution time is dependent on the amount of measurement uncertainty, with a 50% to 85% reduction in execution time over an SNR range of 20 dB. This contrasts with Infomax's execution time, which is unaffected by measurement uncertainty.

Figures (9)

• Figure 1: The experimental setup for our analysis.The Keithley 3390 arbitrary waveform generator is directly coupled to the input pins of the OpenBCI Cyton Board through the use of crocodile clips. The data is transmitted via an RF connection to a USB dongle plugged into the computer. The OpenBCI GUI allows the user to view the data measurement in real time and save off data recordings.
• Figure 2: The distribution of a measurement recording of an approximately constant signal. Each of the histogram bins has a width of approximately $3.4$ µ V. While the voltage of the true signal can be considered constant for the data recording, the measurement distribution has an apparent Gaussian shape.
• Figure 3: A comparison between the measurement record's CDF and the CDF of a Gaussian with the same mean and variance. The CDFs are almost exactly aligned, furthering the evidence that we can consider the measurement uncertainty to be Gaussian in nature.
• Figure 4: The electrode locations of the EEG channels present in the data. The orange electrodes were considered measurement electrodes, and the blue electrode (M2) was used as the reference electrode.
• Figure 5: The four additional electrode configurations used in our simulations. The orange electrodes are measurement electrodes and the blue electrode is a reference electrode. The white electrodes are unused in a particular configuration. (a) Electrodes corresponding to the Biopac Systems B-Alert X10 BCI, the configuration we refer to as "com9" (b) Electrodes selected to optimize for emotion-specialized tasks, the configuration we refer to as "em8". (c) Electrodes selected to optimize for attention-specialized tasks, the configuration we refer to as "att8". (d) Electrodes selected to optimize for motor imagery tasks, the configuration we refer to as "mi10".