Versatile MRI acquisition and processing protocol for population‐based neuroimaging

Abstract Background Neuroimaging plays an essential role in epidemiological studies and a special focus is directed towards the employment of versatile MRI acquisition and processing. The proposed multi‐purpose MRI protocol was designed for large‐scale and long‐term population neuroimaging and includes structural, diffusion‐weighted, and functional MRI modalities. It directly links the acquisition of an extensive set of MRI contrasts with fully automated data processing pipelines and quality assurance of the MRI data and image‐derived phenotypes. Method The MRI acquisition protocol is largely based on in‐house developed MR sequences. With a total scan time below one hour per participant, it allows to acquire multiple MR contrasts at 3T with whole‐brain coverage and high isotropic image resolution, while keeping potential subject discomfort and motion artifacts at manageable levels. Designed to be kept constant, but also adaptive, the scan protocol separates into a core imaging protocol with MR contrasts of high relevance acquired for all study participants, and a free protocol to accommodate alternative promising MRI techniques in smaller sub‐populations. The analysis protocol handles large‐scale imaging data by means of fast and fully automated processing pipelines that incorporate state‐of‐the‐art image analysis tools and innovative machine learning methods, particularly using deep learning. Dedicated quality assessment (QA) includes visual rating and inspection for incidental findings on structural MRI and QA workflows tailored for each postprocessing pipeline to automatically identify problematic data based on the distribution of subject‐specific QA metrics with respect to the population average. Result The MRI protocol has been successfully applied in the Rhineland Study, a prospective cohort study in Bonn, Germany, with currently over 10,000 participants. Image‐derived phenotypes include: global and regional brain tissue volume, thickness and surface measures from multi‐modal structural MRI, global and regional microstructural measures based on diffusion‐weighted MRI, brain functional connectivity using resting‐state functional MRI, and volumes of subcutaneous and visceral adipose tissue based on a single‐breathhold abdominal MRI (Figure 1). Conclusion We present a versatile MRI protocol with acquisition and analysis methods that are generally applicable and not geared towards a specific disease or research question. Thus, this protocol may be of specific interest for many neuroimaging applications including population imaging studies.