Quantitative MRI provides markers of intra-, inter-regional, and age-related differences in young adult cortical microstructure

Caprini, Francesco and Carey, Daniel

Birkbeck Schools and Research Centres > School of Science > Psychological Sciences

Participants were 93 right-handed healthy adults (mean age ± SD: 23.6 ± 4.3; range: 18-39; 57 female, 36 male). All scanning took place at the Wellcome Trust Centre for Neuroimaging (WTCN), London.
Participants were sampled over approximately 24 months. Thirty-four participants were recruited as part of a study of musicianship and consisted of expert violinists (n = 18; mean age ± SD: 22.8 ± 2.8; 13 female, 5 male) and closely matched non-musicians (n = 16; mean age ± SD: 23.3 ± 3.1; 12 female, 4 male). All had completed or were enrolled in a university degree, and were recruited from the University of London, music conservatories in London, and local participant pools.
The remaining participants (n = 59; mean age ± SD: 23.9 ± 4.9; 32 female, 27 male) were sampled from the general population through local participant pools.

The multi-parameter mapping protocol data (Weiskopf et al., 2013; Lutti et al., 2010, 2012) were acquired at the WTCN using a 3T whole-body Tim Trio system (Siemens Healthcare) with radiofrequency body coil for transmission and a 32-channel head coil for signal reception. The MPM protocol consisted of three differently weighted 3D multi-echo FLASH acquisitions acquired with 800 micron isotropic resolution. Volumes were acquired with magnetization transfer (MTw), T1-(T1w), and proton density (PDw) weighting. The MT weighting was achieved through application of a Gaussian RF pulse (4ms duration, 220° nominal flip angle) applied 2kHz off-resonance prior to non-selective excitation.
Two further scans were collected to estimate participant-specific inhomogeneities in the RF transmit field (B1+) using a 3D EPI acquisition of spin-echo (SE) and stimulated echo (STE) images as described in Lutti et al. (2010) (slice thickness: 4 mm; matrix size: 64 x 48 x 48; field-of-view: 256 x 192 x 192 mm3; bandwidth: 2298 Hz/pixel; SE/STE acquisition time post-excitation: 39.38 ms/72.62 ms; TR: 500 ms). In addition, a map of the B0 field was acquired and used to correct the B1+ map for off-resonance effects (Lutti et al., 2010; see also Weiskopf et al., 2006) (voxel size: 3 x 3 x 2 mm3; slice thickness: 4mm; field-of-view: 192 x 192 mm2; 64 slices, 1mm gap; bandwidth: 260 Hz/pixel; TE1 10 ms, TE2 12.46 ms; TR: 1020 ms; flip angle: 90°).

Participants provided written informed consent and were screened for contraindications for MRI. B1+ and B0 field maps were collected at the beginning of each session, followed by the MT, PDw, and T1w scans. Participants’ eye and head movements were monitored using an eye tracker (Eyelink 1000 Core System) during scanning runs. Rest breaks of several minutes were provided between scans as required.

Images were pre-processed using the Voxel Based Quantification (VBQ) toolbox in SPM 8. In brief, regression of the log signal from the echoes of all weighted volumes were used to calculate a map of R2* using the ordinary least squares ESTATICS approach (Weiskopf et al., 2014). The set of echoes for each acquired weighting were then averaged to increase the signal-to-noise ratio (Helms and Dechent, 2009). This was done using only the first six echoes for Cohort 2. The 3 resulting volumes were used to calculate MT, R1, and PD* maps as described in Helms et al. (2008a, 2008b) and Weiskopf et al. (2013). Quantitative R1 values at each voxel were estimated based on the rational approximation of the Ernst equation described by Helms et al. (2008a). To maximize the accuracy of the R1 map, these maps were corrected for transmit field inhomogeneities by constructing a map from the calibration data according to the procedure detailed in Lutti et al. (2012). The R1 maps were also corrected for imperfect spoiling characteristics using the approach described by Preibisch and Deichmann (2009). The MT map was constructed using the procedure described in Helms et al. (2008b). This is a semi-quantitative metric depicting the percentage loss of magnetization resulting from the MT pre-pulse used and differs from the commonly used MT ratio (percentage reduction in steady state signal) by explicitly accounting for spatially varying T1 relaxation times and flip angles (Weiskopf et al., 2013). Finally, PD* maps were estimated from the signal amplitude maps by adjusting for receive sensitivity differences using a post-processing method similar to UNICORT (Weiskopf et al., 2011). To make the PD*maps comparable across participants, they were scaled to ensure that the mean white matter PD* for each subject agreed with the published level of 69% (Tofts, 2003). This quantity is referred to as effective PD (PD*) because it was calculated based on the average FLASH volumes and there was no correction for R2* signal decay.
Following reconstruction of multi-parameter images, all images were manually inspected for any evidence of alignment difficulties, head movement or other image artifacts (e.g., aliasing) by a rater who was blind to subject identity.

Participants’ cortical surfaces were reconstructed using FreeSurfer (v. 5.3; Dale et al., 1999). Use of multi-parameter maps as input to FreeSurfer can lead to localized tissue segmentation failures due to boundaries between the pial surface, dura matter and CSF showing different contrast compared to that assumed within FreeSurfer algorithms (Lutti et al., 2014). Therefore, an in-house FreeSurfer surface reconstruction procedure was developed to overcome these issues. Full details of the processing pipeline are provided in supplemental methods.

Following cortical surface reconstruction, R1, MT, R2* and PD* data were mapped onto participants’ cortical surfaces in FreeSurfer. Whole-brain vertex-wise analyses were subsequently performed.
First, all subjects were rotated to the same (canonical) orientation, using the AFNI 3dwarp routine (-deoblique flag). MPM data were then mapped onto each subject’s surface (using the FreeSurfer mri_vol2surf routine). For each reconstructed hemisphere, quantitative data were sampled along the normal to each surface vertex, for cortical depth fractions from 0.1 (i.e., above white matter surface boundary) to 0.9 (i.e., beneath pial surface boundary) in increments of 0.1 (see Dick et al., 2012).

For each relaxation parameter (R1, MT, R2*, PD*), we first created cross-subject hemisphere-wise average maps for each cortical depth sampling fraction (0.1-0.9) using cortical-surface-based methods with curvature-based alignment (Fischl et al., 1999; Hagler & Sereno, 2006; Dick et al., 2012; Sereno et al., 2013).

The study received approval from the local ethics committee