G. Bartzokis et al., RELIABILITY OF MEDIAL TEMPORAL-LOBE VOLUME MEASUREMENTS USING REFORMATTED 3D IMAGES, PSYCHIATRY RESEARCH-NEUROIMAGING, 82(1), 1998, pp. 11-24
The study assessed whether standardizing the angle of image display an
d controlling for head position in three planes affects the scan-resca
n reliability of medial temporal lobe volume measures when very thin (
1.5 mm) slices are used. Five volunteers were scanned two times oil co
nsecutive days. A three-dimensional MRI sequence acquired whole brain
data in 1.4 mm thick coronal slices. The data were displayed as 1.5 mm
thick images and were rated both in the originally acquired coronal p
lane, and after reformatting to correct for head tilt and display the
brain in the coronal plane perpendicular to the long axis of the left
anterior hippocampus. One rater measured five brain regions (temporal
lobe, anterior and posterior hippocampus, amygdala, and temporal horn)
on the left and right sides of the two non-reformatted and two reform
atted scans to obtain inter-scan variance. Furthermore, most scans wer
e remeasured, to obtain 'reread' variances. All data were log-transfor
med in order to produce comparable variability across brain regions of
different sizes. For all the regions, except the temporal horn, the n
on-reformatted scans showed significantly larger scan-rescan variabili
ty than the reformatted scans. A typical standard deviation for a non-
reformatted pair of scans was 0.10, corresponding to 26% error, while
a typical value for a reformatted pair of scans was 0.04, correspondin
g to 10% error. For all the regions, the reread data (intra-rater reli
ability) gave similar results for both reformatted and non-reformatted
images with similar standard deviations (typical value for reread sta
ndard deviation was 0.020, corresponding to 5% error). The data sugges
t that, even when very thin slices are acquired, volume measurement ac
curacy of gray matter structures in the temporal lobe is considerably
improved by controlling for image orientation in three planes. For the
se structures, the sample size needed to detect a small (5%) within-su
bject volume change would be halved if reformatted images were used. I
mage contrast is an additional important factor since the reformatted
T-1 weighted images used in this study, which have suboptimal CSF/brai
n contrast, worsened measurement accuracy in the temporal horn. (C) 19
98 Elsevier Science Ireland Ltd.