Reliability of medial temporal lobe volume measurements using reformatted 3D images

The study assessed whether standardizing the angle of image display and controlling for head position in three planes affects the scan-rescan reliability of medial temporal lobe volume measures when very thin (1.5 mm) slices are used. Five volunteers were scanned two times oil consecutive 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 plane, 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 reformatted scans to obtain inter-scan variance. Furthermore, most scans were remeasured, to obtain 'reread' variances. All data were log-transformed in order to produce comparable variability across brain regions of different sizes. For all the regions, except the temporal horn, the non-reformatted scans showed significantly larger scan-rescan variability 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, corresponding to 10% error. For all the regions, the reread data (intra-rater reliability) gave similar results for both reformatted and non-reformatted images with similar standard deviations (typical value for reread standard deviation was 0.020, corresponding to 5% error). The data suggest that, even when very thin slices are acquired, volume measurement accuracy of gray matter structures in the temporal lobe is considerably improved by controlling for image orientation in three planes. For these structures, the sample size needed to detect a small (5%) within-subject volume change would be halved if reformatted images were used. Image contrast is an additional important factor since the reformatted T-1 weighted images used in this study, which have suboptimal CSF/brain contrast, worsened measurement accuracy in the temporal horn. (C) 1998 Elsevier Science Ireland Ltd.