MEGAP: A Comprehensive Pipeline for Automatic Preprocessing of Large-Scale Magnetoencephalography Data.

Abstract

Magnetoencephalography (MEG) data are often contaminated by various noise and artifacts, necessitating meticulous preprocessing. However, no pipeline has comprehensively examined all aspects of the different types of MEG noise, nor has any automatic preprocessing pipeline ever been presented. The impracticality of performing visual inspections for the preprocessing of large-scale resting state datasets, combined with the absence of automation, hinders the ability to take advantage of such datasets, including increased generalizability. Additionally, the absence of a standardized sequence for MEG preprocessing steps affects the reproducibility of research studies. Our MEG Automatic Pipeline (MEGAP) can automatically reduce noise and artifacts and is the first pipeline that can be used to preprocess large-scale resting state MEG datasets. We developed this pipeline by integrating and sequencing recent algorithms for each preprocessing step, ensuring automated execution, standardization, and organized outputs. The key features of MEGAP include correcting head movements, removing line noise without applying a notch filter, annotating muscle artifacts, removing sensor and environmental noise, and automatically detecting artifact components in Independent Component Analysis (ICA). We validated our pipeline using simulated and experimental data from the Cambridge Centre for Aging and Neuroscience (Cam-CAN) dataset. Substantial improvements were achieved based on different evaluation criteria such as Normalized Mean Square Error (NMSE), correlation, and Signal to Noise Ratio (SNR). MEGAP provides a robust framework for MEG data preprocessing, significantly reducing the manual effort in preprocessing by automating the required steps, contributing to more consistent and reproducible neuroimaging research outcomes, and facilitating the analysis of large-scale MEG studies.