Premium
IC‐P‐044: Automated and robust algorithm to measure changes in medial temporal lobe volume in early Alzheimer's disease
Author(s) -
Kazemifar Samaneh,
Drozd John,
Rajakumar Nagalingam,
Borrie Michael,
Bartha Robert
Publication year - 2012
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1016/j.jalz.2012.05.076
Subject(s) - segmentation , region of interest , magnetic resonance imaging , artificial intelligence , imaging phantom , temporal lobe , neuroimaging , brain size , white matter , image segmentation , computer science , pattern recognition (psychology) , nuclear medicine , medicine , psychology , radiology , neuroscience , epilepsy
tion in a population of interest. Using pair-wise measures between various features of each subject, manifold learning maps these features to a lower dimensional space, where they can be used for further analysis or classification. In this work, we build a joint PiB/MRI manifold to investigate longitudinal changes in Alzheimer’s disease. Methods: Longitudinal PiB and T1W MRI images from the AIBL study were used (16 AD, 26 MCI and 138 NC). MR images were segmented, skull-striped and normalised to a canonical brain using a soft non-rigid registration. The segmentations and PiB images were then propagated to the canonical space. A cerebellummask defined on the population atlas and masked by the GM segmentation was used to SUVR normalise each PiB images. To remove contrast variations due to differing MR scanners a composite MR image was generated from the GM, WM and CSF segmentations. Pair-wise Normalised Mutual Information was computed on the PiB and composite MR images, and the 2 affinity matrices summed to produce a joint PiB/MRI manifold embedding using Laplacian eigen-maps. Results: The resulting embedding exhibits two preferential axes, with left-to-right increases the PiB retention, and top-tobottom increases in atrophy. (Figure 1) Two preferential paths representing the evolution of the subjects in the cohort can be identified. One path (Fig 1 left) contains mostly NC and some MCI which don’t show any PiB increase over time, but exhibit higher atrophy, which presumably correspond to subjects undergoing either normal ageing or non-Ab related neurodegenerative process. The second path encompass a wider range of clinical classifications where subjects progress to the right as Ab accumulates followed by a downward trend as atrophy increases. Conclusions: An unbiased PiB/MRI population model was developed to analyse the pattern of change in Ab accumulation and brain volume over time. According to this model, the transition fromPiBtoPiB+appears to be independent of atrophy (but not age),which is consistent with a model of Ab deposition preceding neurodegeneration.