NeuroClean: multipurpose neural data preprocessing pipeline

Abstract

[en] Electroencephalography (EEG) and Local field potentials (LFP) are two commonly used measures of electrical activity in the brain. These signals are used extensively in both industry and research and have many real world applications. Before any analyses can be performed on EEG/LFP, however, the data must first be cleaned. The main objective of this project was to create an unsupervised, multipurpose EEG/LFP preprocessing pipeline. Its unsupervised nature would, consequently, help alleviate problems involving reproducibility and biases that arise from human intervention. Moreover, manual signal cleaning is time and labor intensive. The adoption of an automated workflow would, therefore, save researchers valuable time and resources. A secondary goal was to allow the pipeline to be fit to several use cases, thus standardizing the cleaning methods used in neuroscience. We designed an automated EEG/LFP preprocessing pipeline, NeuroClean, which consists of five steps: bandpass filtering, line noise filtering, bad channel rejection, and independent component analysis with automatic component rejection based on a clustering algorithm. Machine learning classifiers were used to ensure task-relevant signals were preserved after each step of the cleaning process. We used an LFP dataset recorded from a cynomolgus macaque to validate the pipeline. Data was recorded while the monkey performed a reach-to-grasp task, and three sections of the movement were used for classification. NeuroClean appeared to remove several common types of artifacts from the signal. Moreover, it yielded over 97% accuracy (whereas chance-level is 33.3%) in an optimized Multinomial Logistic Regression model after cleaning the data, compared to the raw data which performed at 74% accuracy. The results show that NeuroClean is a promising pipeline and workflow that may be explored in the future.