Yet another artefact rejection study: an exploration of cleaning methods for biological and neuromodulatory noise

Objective. Electroencephalography (EEG) cleaning has been a longstanding issue in the research community. In recent times, huge leaps have been made in the field, resulting in very promising techniques to address the issue. The most widespread ones rely on a family of mathematical methods known as blind source separation (BSS), ideally capable of separating artefactual signals from the brain originated ones. However, corruption of EEG data still remains a problem, especially in real life scenario where a mixture of artefact components affects the signal and thus correctly choosing the correct cleaning procedure can be non trivial. Our aim is here to evaluate and score the plethora of available BSS-based cleaning methods, providing an overview of their advantages and downsides and of their best field of application. Approach. To address this, we here first characterized and modeled different types of artefact, i.e. arising from muscular or blinking activity as well as from transcranial alternate current stimulation. We then tested and scored several BSS-based cleaning procedures on semi-synthetic datasets corrupted by the previously modeled noise sources. Finally, we built a lifelike dataset affected by many artefactual components. We tested an iterative multistep approach combining different BSS steps, aimed at sequentially removing each specific artefactual component. Main results. We did not find an overall best method, as different scenarios require different approaches. We therefore provided an overview of the performance in terms of both reconstruction accuracy and computational burden of each method in different use cases. Significance. Our work provides insightful guidelines for signal cleaning procedures in the EEG related field.