Development of χ-separation pipeline for UK Biobank dataset

Motivation: Recently developed χ-sepnet-$$$R_2 ^*$$$ provides opportunities to apply χ-separation in the UK Biobank (UKB) dataset. However, poor quality $$$R_2 ^*$$$ from the 3 mm-thick UKB GRE data hampers to create high-quality χ-separation results. Goal(s): Our goal is to develop a pipeline for generating high-quality χ-separation maps from the UKB dataset. Approach: We developed a new neural network that improved $$$R_2 ^*$$$ quality of the UKB data. Additionally, a full data processing pipeline that utilized χ-sepnet-$$$R_2 ^*$$$ for the UKB data was proposed. Results: $$$R_2 ^*$$$ network improved the quality of $$$R_2 ^*$$$, and the proposed pipeline showed successful susceptibility source separation outcomes. Impact: This study proposes a processing pipeline for high-quality χ-separation in the UKB dataset. For high-quality χ-separation, B0-field inhomogeneity artifact in $$$R_2 ^*$$$ was removed using a neural network. Our pipeline enables us to investigate the large cohort UKB data.