Slice-to-Volume Registration Enables Automated Pancreas MRI Quantification in UK Biobank

Multiparametric MRI of the pancreas can potentially benefit from the fusion of multiple acquisitions. However, its small, irregular structure often results in poor organ alignment between acquisitions, potentially leading to inaccurate quantification. Recent studies using UK Biobank data have proposed using pancreas segmentation from a 3D volumetric scan to extract a region of interest in 2D quantitative maps. A limitation of these studies is that potential misalignment between the volumetric and single-slice scans has not been considered. In this paper, we report a slice-to-volume registration (SVR) method with multimodal similarity that aligns the UK Biobank pancreatic 3D acquisitions with the 2D acquisitions, leading to more accurate downstream quantification of an individual’s pancreas T1. We validate the SVR method on a challenging UK Biobank subset of N = 50, using both direct and indirect performance metrics.