PELICAN: a Longitudinal Image Processing Pipeline for Analyzing Structural Magnetic Resonance Images in Aging and Neurodegenerative Disease Populations

Abstract

Structural magnetic resonance imaging (MRI) allows for accurate non-invasive assessment of the brains structure and its longitudinal changes. Availability of large scale longitudinal MRI datasets enables us to probe brain changes in health and disease, and derive longitudinal trajectories of brain morphometry based measures to estimate brain atrophy and other disease-related abnormalities. In contrast to their cross-sectional counterparts, image processing pipelines that have been designed for longitudinal data can reduce noise in the derived measurements by disentangling the within and between subject variabilities, improving the sensitivity of the downstream models in detecting more subtle longitudinal changes. Here we present PELICAN, our open source multi-contrast longitudinal image processing pipeline, that has been designed and extensively validated for use in longitudinal settings and populations with neurodegenerative disorders. PELICAN can use population specific average templates as intermediate targets to derive accurate nonlinear registrations for cases with substantial levels of atrophy, which commonly used pipelines struggle to process. We evaluated PELICANs performance across over 34,000 MRIs from multiple aging and neurodegenerative disorder cohorts, and compared its reliability and failure rates against FreeSurfer as a widely used image processing tool, showing superior performance of PELICAN compared to FreeSurfer, both in terms of failure rate and reliability. Our results demonstrate that PELICAN can be used to accurately process MRIs of individuals with neurodegenerative disease who present with greater levels of atrophy and white matter lesion burden.