Lowe VJ, Weigand SD, Jack Jr CR, Kantarci K, Senjem ML, Petersen RC.
Optimized ROI measures of FDG PET improve characterization of early cognitive impairment and are comparable to PIB PET.
Human Amyloid Imaging 2011 Meeting Abstracts. 2011 Jan 15;
Objective: PiB or FDG PET may be most helpful when characterizing early cognitive impairment. Improvements
in FDG PET data analysis as compared to PiB may be possible by using selective, strategic ROI measurements
as hypometabolism occurs selectively early in dementia. We attempted to optimize FDG PET performance as
compared to PiB PET and utilize it to discriminate amnestic (aMCI), non amnestic MCI (naMCI) and AD.
Statistical methods: For FDG PET of 96 ADNI subjects with AD and 104 subjects who were cognitively normal
(CN), we fit a penalized logistic regression model with AD as the event and the pons-normalized uptake values
for 46 bilateral ROIs as the predictors. We selected the model using cross-validation of the area under the ROC
curve (AUROC). We used the resulting model to calculate an estimated probability of aMCI, naMCI, or AD by FDG
as compared to PiB PET for subjects in our Mayo Clinic cohort that consisted of CN (340), aMCI (91), naMCI (21)
and AD (31) subjects.
Results: FDG ROIs that contributed most, and with regression coefficients that were approximately equal in
magnitude, were posterior cingulum (PC), angular, and the postcentral gyrus (PG). Reduced uptake in the PC and
angular ROIs was associated with increased odds of AD while increased uptake in the PG was associated with
increased odds of AD, after holding all other ROIs constant. These ROIs were then used as an goptimizedh FDG
measure. The AUROC (95% CI) for PiB, global FDG and optimized FDG PET for CN vs. aMCI was 0.72 (0.66, 0.78),
0.61 (0.54, 0.67), and 0.71 (0.65, 0.77). PiB and optimized FDG performed significantly better than global FDG. For
aMCI vs. naMCI, the values were: 0.73 (0.59, 0.83), 0.53 (0.40, 0.66), and 0.66 (0.52, 0.77) with PiB and optimized
FDG both performing better than global FDG. Similar discriminative performance of the PET models was seen
when AD was the comparative group.
Conclusion: Our regression model identified an optimized FDG PET analysis method that can help better
discriminate between CN, aMCI, and naMCI with discrimination comparable to that of PiB PET. This analysis
method may improve the characterization of early cognitive impairment.