Reproducibility and Reliability of Quantitative and Weighted T1 and T2(∗) Mapping for Myelin-Based Cortical Parcellation at 7 Tesla.

Different magnetic resonance (MR) parameters, such as R1 (=1/T1) or T2(∗), have been used to visualize non-invasively the myelin distribution across the cortical sheet. Myelin contrast is consistently enhanced in the primary sensory and some higher order cortical areas (such as MT or the cingulate cortex), which renders it suitable for subject-specific anatomical cortical parcellation. However, no systematic comparison has been performed between the previously proposed MR parameters, i.e., the longitudinal and transversal relaxation values (or their ratios), for myelin mapping at 7 Tesla. In addition, usually these MR parameters are acquired in a non-quantitative manner ("weighted" parameters). Here, we evaluated the differences in 'parcellability,' contrast-to-noise ratio (CNR) and inter- and intra-subject variability and reproducibility, respectively, between high-resolution cortical surface maps based on these weighted MR parameters and their quantitative counterparts in ten healthy subjects. All parameters were obtained in a similar acquisition time and possible transmit- or receive-biases were removed during post-processing. It was found that CNR per unit time and parcellability were lower for the transversal compared to the longitudinal relaxation parameters. Further, quantitative R1 was characterized by the lowest inter- and intra-subject coefficient of variation (5.53 and 1.63%, respectively), making R1 a better parameter to map the myelin distribution compared to the other parameters. Moreover, quantitative MRI approaches offer the advantage of absolute rather than relative characterization of the underlying biochemical composition of the tissue, allowing more reliable comparison within subjects and between healthy subjects and patients. Finally, we explored two parcellation methods (thresholding the MR parameter values vs. surface gradients of these values) to determine areal borders based on the cortical surface pattern. It is shown that both methods are partially observer-dependent, needing manual interaction (i.e., choice of threshold or connecting high gradient values) to provide unambiguous borders.