Whole brain structural changes predict the mild cognitive impairment to Alzheimer's disease conversion: Evidence from the Alzheimer's Disease Neuroimaging Initiative

Abstract

approximately 90% of the total variation are used as features. Hippocampal volume corrected for total intracranial volume is added as independent feature. The shape components are evaluated in the AD classification using bagged SVM with 25 resamplings. Results: Using hippocampal volume gave on average 83.47% accuracy (average of 20 repetitions). sing the features produced by SSMs alone provided the best performance (87.36%) when computed on the landmarks selected at p<0.05 (see Table). When combining hippocampus volume with the shape features, the best classification performance (88.07%) is achieved using the subregion identified by p<0.1. The Hotelling’s T2 significance map on the hippocampus shows that the regions of highest discrimination match with the CA1 and subiculum subfields (see Figure). Conclusions: Using hippocampal shape descriptors provides extra discrimination between AD and NC in addition to using volume alone. The shape components on the regions exhibiting more significant effect of the atrophy indicate a better representation of the hippocampal shape change associated with AD.