Can the performance of the FSL/Siena whole brain atrophy measure be improved at 3T and 1.5T on ADNI back-to-back MPRAGES by using a non-default settings?

Background: To test a previous report of improved performance of the FSL/ Siena whole brain atrophy algorithm by using non default settings at both 3T and 1.5T. Methods: As part of the Alzheimer's Disease Neuroimaging Initiative (ADNI) study, high quality MPRAGE (a T1 weighted 3D MRI sequence) from 200 subjects visit pairs is available to compare the reproducibility of brain atrophies over 12 to 18 month intervals at both 3T and 1.5T. This is because in ADNI all subjects received back-to-back (BTB) MPRAGEs with the second MRPAGE acquisition starting within seconds of the completion of the first. Also as part of ADNI, 131 subjects received BTB scans at multiple time points and two field strengths of 3T and 1.5T. For this study FSL/Siena was applied to the 200 subject visit pairs for 88 different combinations of options and the two field strengths requiring a total of 9 core years of calculations. For each combination of options and field strengths Siena's BTB reproducibility was quantified as the median of the absolute value of the difference between the BTB percentage brain volume change (PBVC) yielded by Siena. Results: For Siena option combinations -B (bias field correction and neck cleanup turned on) and f = 0.1 to 0.3 (which selects the fractional intensity threshold) provided the smallest reproducibility of 0.26 to 0.27 for 1.5T. However, bootstrapping provided a 95% confidence interval on the reproducibility up to 0.29. Options that often yielded poor reproducibilities included the -S option (eye and optic nerve cleanup turned on) and f > = 0.5. The range of the 3T results made an optimal option combination less clear. For the 69 subjects with two visit pairs each, no significant correlation was found between the BTB differences of the 0 to 12 month and 6 to 18 month visit pairs bearing out their treatment in this analysis as independent measures. Conclusions: For the ADNI MPRAGEs at 1.5T the -B option with f = 0.3 will likely yield the best BTB PBVC reproducibility for Siena - supporting the previous report and differing from the Siena default value of f = 0.5. For 3T the optimum choice is less clear.