Paul Dufort PhD, Presenter: Nothing to Disclose

Adrian P. Crawley PhD, Abstract Co-Author: Nothing to Disclose

David John Mikulis MD, Abstract Co-Author: Stockholder, Thornhill Research Inc Research Grant, General Electric Company

To determine if cerebrovascular reactivity (CVR) is a potential metric for distinguishing Alzheimer’s Disease (AD) from Mild Cognitive Impairment (MCI) patients and age-matched controls (NC).

Anatomical and CVR BOLD MRI scans were performed in 5 NC, 5 AD, and 6 MCI subjects. The subjects underwent two 10 mmHg prospective iso-oxic square wave increases in end-tidal CO2 lasting 45 sec. and 130 sec. with an intervening 90 sec. normocapnic period. The regression coefficient of the % change in BOLD MRI signal vs. mmHg change in end-tidal CO2 is the CVR metric. CVR maps and anatomical scans were normalized to MNI152 space using SPM 8. A sparse logistic binomial regression classifier (GLMnet) was then trained to differentiate between NC and AD subjects only, using as input: (i) the set of CVR values at each spatially standardized grey matter (GM) voxel; or (ii) the statistical quantiles of texture features representing spatial heterogeneity in the CVR maps over all GM voxels. As an additional quantification of differences in spatial heterogeneity, the distribution of power in each spatial frequency band was calculated and displayed.

The classifier based on CVR at each GM voxel performed with 80%/100% sensitivity and specificity under leave-one-out cross-validation, and was not significant (p = 0.34) under random label permutation. The classifier based on CVR GM texture features had 100% sensitivity and specificity (when trained on any 9 of the subjects and tested on the tenth, all test cases were classified correctly), and a significance of p = 0.007 under random label permutation. Figure 1a shows the probability of AD as a function of group for the texture-based classifier, while Figure 1b shows the differences in spatial power distribution.

Despite the small sample size and the exclusion of the MCI patients in training, the classifier found a classification axis placing the MCIs between the NCs and AD subjects based only on CVR texture features. The classifier may identify progression toward AD based on changes in CVR before either cognitive decline or gray matter loss, warranting a prospective assessment of efficacy.

The MRI CVR texture metric that has been developed could become an efficient and effective alternative means for screening patients with AD.  

Dufort, P, Crawley, A, Mikulis, D, Cerebrovascular Reactivity Can Distinguish Alzheimer’s Disease from Patients with Mild Cognitive Impairment, and Age Matched Controls.  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14015033.html