Reproducibility and accuracy of quantitative magnetic resonance imaging techniques of whole-brain atrophy measurement in multiple sclerosis

Background and Purpose. Whole-brain atrophy is of growing interest as an outcome measure in multiple sclerosis (MS) clinical trials. The authors compared the reproducibility and accuracy of 3 quantitative techniques of measurement in patients with MS. Methods. Thirty-four patients with relapsing-remitting MS (median Expanded Disability Status Scale disability score = 1.5) were studied. Brain parenchymal fraction (BPF) was quantified on spin-echo 2-dimensional T1-weighted axial 5-mm slice thickness sequences by semiautomated (Buffalo, Trieste) or automated (SIENAX algorithms. Results. Mean +/- SD BPFs were 0.830 +/- 0.04 with Buffalo, 0.824 +/- 0.04 with Trieste, and 0.826 +/- 0.04 with SIENAX methods (P nonsignificant [NS]). Mean BPF scan-rescan coefficient of variation (COV) was 0.41% for Buffalo, 0.44% for Trieste, and 0.32% for SIENAX (P = NS). The semiautomated methods showed higher accuracy than the automated method in brain extraction (masking; P = .001). The errors of skull stripping included scalp, skull bone marrow, inferior parts of temporal lobes anterior to the brain stem, face structures, sagittal sinuses, eyes, and optic nerves. Buffalo (r = -0.37, P = .034) and Trieste (r = -.36, P = .039) BPFs showed stronger correlation with disability than SIENAX (r = -0.16, P = .219). These differences Were statistically significant (P = .0031 for Buffalo and P -.0037 for Trieste BPF). Conclusions. This study showed a high reproducibility of both semiautomated and automated methods for brain atrophy measurement. The semiautomated methods showed higher accuracy than the automated SIENAX method did in the evaluation of brain extraction, especially in infratentorial and cortical regions, where operator interaction during the masking processes was essential.