Comparison of chi-separation and chi-separation*: A clinical feasibility study

While χ-separation gives unique and valuable information, requirement of R2 limited its clinical application. This study assessed the clinical feasibility of two R2*-based magnetic source separation. From 24 MR scans of MS subjects, four sets of magnetic source separation maps were generated; (1) χ-separation (originally proposed method), (2) χ-separation* (exploiting R2*), (3) χ-sepnet (DL-based), and (4) χ-sepnet* (DL-based, utilizing R2*). χ-separation* and χ-sepnet* showed good agreement to χ-separation and χ-sepnet in MS lesions and the brain. However, there are small but consistent biases in each method, which are important caveats for R2*-based χ-separation. These biases were relatively smaller in χ-sepnet*.