Adjusting ADJUST: Optimizing the ADJUST algorithm for pediatric data using geodesic nets

Abstract

A major challenge for electroencephalograph (EEG) studies on pediatric populations is that large amounts of data are lost due to artifacts (e.g., movement and blinks). Independent component analysis (ICA) can separate artifactual and neural activity, allowing researchers to remove such artifactual activity and retain a greater percentage of EEG data for analyses. However, manual identification of artifactual components is time‐consuming and requires subjective judgment. Automated algorithms, like ADJUST and ICLabel, have been validated on adults, but to our knowledge, no such algorithms have been optimized for pediatric data. Therefore, in an attempt to automate artifact selection for pediatric data collected with geodesic nets, we modified ADJUST's algorithm. Our “adjusted‐ADJUST” algorithm was compared to the “original‐ADJUST” algorithm and ICLabel in adults, children, and infants on three different performance measures: respective classification agreement with expert coders, the number of trials retained following artifact removal, and the reliability of the EEG signal after preprocessing with each algorithm. Overall, the adjusted‐ADJUST algorithm performed better than the original‐ADJUST algorithm and no ICA correction with adult and pediatric data. Moreover, in some measures, it performed better than ICLabel for pediatric data. These results indicate that optimizing existing algorithms improves artifact classification and retains more trials, potentially facilitating EEG studies with pediatric populations. Adjusted‐ADJUST is freely available under the terms of the GNU General Public License at: https://github.com/ChildDevLab/MADE-EEG-preprocessing-pipeline/tree/master/adjusted_adjust_scripts.

Automated Pipeline for Infants Continuous EEG (APICE): a flexible pipeline for developmental studies

“…The copyright holder for this preprint (which this version posted May 23, 2021. ; https://doi.org/10.1101/2021.05.21.445085 doi: bioRxiv preprint thus, it is possible that other available methods, like adjusted-ADJUST (Leach et al, 2020), may provide better performance. One possible explanation is that these methods are not optimal for infant data due to the intrinsic properties of the signal.…”

“…Recently, some algorithms have been adapted for infant data. For example, the ADJUST algorithm (Mognon et al, 2011) has been adapted into the adjusted-ADJUTS (Leach et al, 2020), and the MARA algorithm (Winkler et al, 2011) into iMARA (Marriot Haresign et al, 2021). While these modified algorithms work better on infant data, they rely on optimizations done on adult training sets.…”

The influence of monetary reward on proactive and reactive control in adolescent males

“…ICA weights from the ICA run on the copied (1 Hz) dataset were then copied back to the continuous 0.3 Hz high-passed data. The adjusted-ADJUST Matlab scripts ( Leach et al, 2020 ; Mognon et al, 2011 ) identified artifactual independent components, which were then removed from the data. The data were epoched from −1000 ms before to 2000 ms after both the cue and the stimulus.…”

Behavioral Inhibition and Dual Mechanisms of Anxiety Risk: Disentangling Neural Correlates of Proactive and Reactive Control

“…During data collection, electrodes were referenced to electrode Cz. All pre-processing, including ocular artifact detection and removal, was performed with the Maryland Analysis of Developmental EEG (MADE) pipeline (Debnath et al, 2020), which utilizes MATLAB (The MathWorks, Natick, MA) functions from EEGLAB (Delorme & Makeig, 2004) and its plugins “FASTER” (Nolan et al, 2010), “ADJUST” (Mognon et al, 2011), and “ADJUSTED ADJUST” (Leach et al, 2020). Offline, data were re-referenced to an average reference and band-pass filtered from 0.3 to 50 Hz with a digital FIR filter.…”