Improved Detection and Mapping of Dynamic Hippocampal and Ventricular Change in Alzheimer’s Disease, Using 4D Parametric Mesh Skeletonization

Paul M. Thompson1, Kiralee M. Hayashi1, Greig de Zubicaray2, Andrew L. Janke2, Stephen E. Rose2, James Semple3, Michael S. Hong1, David H. Herman1, David Gravano1 Stephanie Dittmer1, David M. Doddrell2, Arthur W. Toga1
1Laboratory of Neuro Imaging, Brain Mapping Division, and UCLA Alzheimer Disease Center, Department of Neurology, UCLA School of Medicine
2Centre for Magnetic Resonance, University of Queensland, Brisbane 4072, Australia
3GlaxoSmithKline Pharmaceuticals plc, Cambridge, UK

We developed an anatomical mapping technique that was found to be more sensitive than volumetric measures for detecting hippocampal and ventricular changes in Alzheimer’s disease (AD).

Fig. 1: Hippocampal and Ventricular Models in Alzheimer's Disease. (a) 3D hippocampal surface; (b) close-up of surface parameterization; (c) statistical maps linking hippocampal shape and diagnosis (c); ventricular expansion and MMSE scores (d).

Methods. 3D anatomical models of the hippocampus and ventricles were derived from a longitudinal database of MRI scans of 12 AD patients and 14 matched controls (ages: 68.4+/-1.9SE and 71.4+/-0.4 yrs.; interscan interval: 2.1+/-0.4 yrs.). 3D parametric mesh models of these structures were registered and skeletonized (i.e., a 3D medial curve was derived). Distance fields relating surface boundary points to their medial curves were analyzed using surface-based nonparametric regression. Permutation tests, on suprathreshold clusters of surface statistics, assessed effects of diagnosis, time, hemisphere, and linkages with declining cognition (as MMSE scores worsened from 17.7+/-1.9 to 12.9+/-2.5 in patients; p<0.0005).

Results. In AD, greatest dynamic change rates were found in the inferior ventricular horns (L:+18.1+/-3.8 %/yr.; R:+12.8+/-4.7%/yr.); significant expansion rates were found bilaterally even in controls (L:+3.7+/-1.1%/yr.; R:+1.7+/-1.1%/yr.; p<0.001,p<0.01). Hippocampal loss rates were faster in AD than controls (L:-4.9+/-1.8%/yr. vs. –3.8+/-1.6%/yr., R: -8.2+/-2.6%/yr. vs. –0.2+/-1.2%/yr.; p<0.01). For both structures, shape abnormalities linked with lower MMSE scores. By orthogonalizing out non-disease-specific variance [1], radial distance maps to 3D medial curves detected diagnostic and cognitive effects more powerfully than volume measures (typically p<0.0001 rather than p<0.05 for both structures). With either volumes or maps, ventricular effects were most powerfully detected. The left hippocampal head showed severe deficits (c). At baseline, maps of ventricular expansion (d) were linked with diagnosis (L:p<0.002, R:p<0.047) and MMSE (L:p<0.002, R:p<0.007). These effects intensified at follow-up. The most powerful measure was a map of the ratio of the hippocampal radial map to the ventricular radial map, averaged across corresponding parametric surface locations: both this map (at baseline and follow-up) and its annualized rate of change, linked strongly with diagnosis and MMSE (all p’s<0.003). Consistent with our cortical studies that mapped a dynamically spreading wave of gray matter loss [2], the left hemisphere was worst affected at this stage of AD, and all deficits were progressive.

Conclusion. These dynamic maps chart how AD progresses. Their increased sensitivity may empower drug and genetic studies of dementia.

References: [1]. Styner M, Gerig G, CVPR, 2001. [2] Thompson PM et al., Journal of Neuroscience, 23(3), Feb. 2003 [in press].