Brain morphometry with multiecho MPRAGE

Abstract

In brain morphometry studies using magnetic resonance imaging, several scans with a range of contrasts are often collected. The images may be locally distorted due to imperfect shimming in regions where magnetic susceptibility changes rapidly, and all scans may not be distorted in the same way. In multispectral studies it is critical that the edges of structures align precisely across all contrasts. The MPRAGE (MPR) sequence has excellent contrast properties for cortical segmentation, while multiecho FLASH (MEF) provides better contrast for segmentation of subcortical structures. Here, a multiecho version of the MPRAGE (MEMPR) is evaluated using SIENA and FreeSurfer. The higher bandwidth of the MEMPR results in reduced distortions that match those of the MEF while the SNR is recovered by combining the echoes. Accurate automatic identification of cortex and thickness estimation is frustrated by the presence of dura adjacent to regions such as the entorhinal cortex. In the typical MPRAGE protocol, dura and cortex are approximately isointense. However, dura has substantially smaller T2* than cortex. This information is represented in the multiple echoes of the MEMPR. An algorithm is described for correcting cortical thickness using T2*. It is shown that with MEMPR, SIENA generates more reliable percentage brain volume changes and FreeSurfer generates more reliable cortical models. The regions where cortical thickness is affected by dura are shown. MEMPR did not substantially improve subcortical segmentations. Since acquisition time is the same for MEMPR as for MPRAGE, and it has better distortion properties and additional T2* information, MEMPR is recommended for morphometry studies.

Figure 1

Partial axial slice through right side of brain showing cortex and adjacent isotense dura (0.35 mm isotropic MPRAGE).

Figure 2

Histogram of ratio image values sampled 0.5 mm to 1 mm out from the gray/white junction. The blue line indicates the μ+2σ threshold. Note the Gaussian shape of the distribution.

Figure 3

a. Comparison of percent brain volume change (PBVC) calculated by SIENA for 12 subjects on 2 scanners for MPR^↑ vs. MPR^↓, MEMPR^↑↓ vs. MEMPR^↓↑ and MEMPR^↑↑ vs. MEMPR^↓↓. b. Comparison of percent brain volume change (PBVC) calculated for 12 subjects using 6 sequence types for one Siemens 3 T TIM Trio vs. another. Note the scaling difference on the y axes.

Figure 4

MPR^↑ (top left), MPR^↓ (bottom left), MEMPR^↑↑ (top middle), MEMPR^↓↓ (bottom middle), MEMPR^↑↓ (top right) and MEMPR^↓↑ (bottom right). All images show white matter surfaces (green) and the two pial surfaces (red) calculated from both the images with opposite readout directions (top vs. bottom rows).

Figure 5

Average displacement in mm between pial surfaces calculated from scans with opposite readout directions, displayed on right hemisphere rotated to show cortex where B0 offsets are greatest. MPR^↑/^↓ (left), MEMPR^↑↑/^↓↓ (middle) and MEMPR^↑↓/^↓↑ (right).

Figure 6

First (top) and fourth (bottom) echo of MEMPR showing difference in T2* contrast between dura and gray matter. In the first echo, as with single echo MPR, gray matter and dura are approximately isointense. All images show white matter surfaces (green) and the pial surface calculated without dura avoidance (yellow) and with dura avoidance (red).

Figure 7

Average cortical thickness differences in mm due to dura correction for MEMPR^↑↑ (left) and MEMPR^↑↓ (right).

Figure 8

Coronal slices through medial temporal susceptibility region (left) and medial frontal susceptibility region (right) showing regions where the first to fourth echo ratio image of the MEMPR exceeds the threshold for dura/non-cortex detection. The pial surface without dura avoidance is shown in yellow and the pial surface with dura avoidance is shown in red. Dura is avoided in the medial temporal susceptibility region, whereas the surface is largely unaffected in the medial frontal susceptibility area where dura does not interfere with cortex.

Figure 9

Volumes of brain structures averaged across subjects and scanners for three sequence types. 1=cerebral white matter, 2=cerebral cortex, 3=lateral ventricle, 4=inferior lateral ventricle, 5=thalamus proper, 6=caudate, 7=putamen, 8=pallidum, 9=hippocampus, 10=amygdala. Volumes are shown for left hemisphere – right hemisphere was similar.

Figure 10

Volume differences between brain structures scanned between the two scanners, averaged across subjects, for three sequence types. Structure numbers follow Figure 9 and are for the left hemisphere. Left and right hemisphere results were similar.

Figure 11

MPR and MEMPR (same, opposite direction readout) SNR values for linear discriminant between adjacent structure pairs. 1=white/gray, 2=white/thal., 3=white/caudate, 4=white/putamen, 5=white/pallidum, 6=white/hippo., 7=white/amygdala, 8=gray/hippo., 9=gray/amygdala, 10=putamen/pallidum, 11=hippo./gray, 12=hippo./amygdala, 13=hippo./lat. ventr., 14=hippo./inf. lat. ventr., 15=amygdala/white, 16=amygdala/gray, 17=amygdala/lat. ventr., 18=amygdala/inf.=lat. ventr., 19=thalamus/caudate, 20=thalamus/lat. ventr., 21=thalamus/pallidum.