Abstract

Alzheimer's disease (AD) is characterised by a decline in cognitive function, and an increase in brain atrophy, the rate of which can be measured by applying registration and the Boundary Shift Integral (BSI) to longitudinal MRI. The initial stage of BSI calculation typically involves time-consuming semi-automated delineation of brain on baseline and repeat scans. Automated propagation of the baseline brain region onto the repeat scan has been suggested as a method to reduce operator time. This study compares the performance, in terms of BSI values, of a novel automated propagation method with semi-automatic segmentation based on BSI values. Twenty-five AD, 74 MCI and 54 control subjects were selected from the Alzheimer's Disease Neuroimaging Initiative cohort, each with a volumetric, T1-weighted 1.5T scan at baseline and 12 months. Scans from both time points were first semi-automatically segmented, with a threshold-based technique. The baseline scan was then registered to the repeat using both affine and freeform registration and the derived registration parameters were used to transform the baseline region to the repeat image. Morphological operations (1 erosion, 2 conditional dilations) were applied to improve boundary location. Finally, whole-brain BSIs were calculated for each subject by registering the semi-automatically segmented repeat and the automatically segmented (propagated) repeat separately to the baseline. Agreement between the resulting semi-automatic and propagated BSIs was calculated by plotting the 95% levels (1.96 std). Limits of agreement were found to be narrow: for 95% of cases, propagated BSIs were between 0.108% greater and 0.128% lower than the semi-automated method (figure). There was a slight but significant bias (p<0.05) with semi-automated BSIs being ∼0.01% of brain volume higher; this is much lower than the BSI scan-rescan reproducibility (0.2%) and very much lower than typical annual brain atrophy in AD (∼2%). For low/negative rates of atrophy the semi-automated BSI tends to be slightly smaller than propagated, and for larger rates the semi-automated BSI tends to be larger. Although there is a small degree of bias the propagation method is comparable to the conventional semi-automated methods in terms of the BSI result and saves operator time.

Full Text
Paper version not known

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call