In the original manuscript, there were two issues with the application of the multi-resolution registration (MRR) step of our analysis, applied to combine three orthogonal ultra-low-field acquisitions with low through-plane resolution (AXI, COR, SAG) into a single higher-resolution scan.First, due to an error in our code (pointed out following its release on GitHub), our MRR analysis involved combination of COR and SAG scans only (i.e., the AXI scan was not included in the ANTs command which is central to executing MRR). Second, to align inputs before running MRR, we pre-registered each ultra-low-field (64 mT) scan to the participant’s corresponding high-field (3 T) scan; however, this raises concerns regarding circular use of high-field data, which is subsequently used as a reference standard.We now repeated all analyses including the AXI scan, using two different pre-registration approaches: the original pre-registration to individual high-field scans, and an alternative pre-registration to a standard high-field template (MNI ICBM152 nonlinear asymmetric), of the same contrast as the input ultra-low-field scans.The impact on results was minor, and is summarised within the following table (an extension of Table 1), depicting summary statistics capturing both 64 mT test–retest reliability, and correspondence to 3 T.Results show that both the number of input scans (2 VS 3) and choice of pre-registration reference (individual 3 T HF scan VS MNI template) have a minimal impact on results. For most contrasts and statistics, median values across 98 SynthSeg+ labels are either identical, or only differ by 0.01-0.02. Only for Lin’s concordance correlation coefficient (CCC) is there a greater decrease in median value (up to 0.08).As such, our corrected results remain consistent with our original conclusions, in that of the two contrasts (T1w and T2w) and five 64 mT scan resolutions investigated (AXI, COR, SAG, ISO, MRR), T2w MRR scans show optimal performance, independent of analytic decisions or errors within the MRR pipeline.As previously, we note that both raw data and (now corrected) code are openly available, enabling replication or extension of our results.
Váša et al. (Thu,) studied this question.