NeuroNorm: An R package to standardize multiple structural MRI

Abstract

Preprocessing of structural MRI involves multiple steps to clean and standardize data before further analysis. Typically, researchers use numerous tools to create tailored preprocessing workflows that adjust to their dataset. This process hinders research reproducibility and transparency. In this paper, we introduce NeuroNorm, a robust and reproducible preprocessing pipeline that addresses the challenges of preparing structural MRI data. NeuroNorm adapts its workflow to the input datasets without manual intervention and uses state-of-the-art methods to guarantee high-standard results. We demonstrate NeuroNorm’s strength by preprocessing hundreds of MRI scans from three different sources with specific parameters on image dimensions, voxel intensity ranges, patients characteristics, acquisition protocols and scanner type. The preprocessed images can be visually and analytically compared to each other as they share the same geometrical and intensity space. NeuroNorm supports clinicians and researchers with a robust, adaptive and comprehensible preprocessing pipeline, increasing and certifying the sensitivity and validity of subsequent analyses. NeuroNorm requires minimal user inputs and interaction, making it a userfriendly set of tools for users with basic programming experience.