Alzheimer's disease pattern derived from pharmacokinetic modeling of ¹¹C-Pittsburgh compound B PET scans

Objectives:
Disease patterns (DP) derived from the scaled subprofile model, a principal component analysis based method (SSM/PCA), have been used to successfully identify Alzheimer’s disease (AD) patients based on ¹⁸F-FDG PET scans. In the current study, a similar approach was explored with parametric images derived from pharmacokinetic analysis of ¹¹C-Pittsbough Compound B (PIB) to generate DPs based on different imaging characteristics of the disease.

Methods:Pharmacokinetic modeling using the simplified reference tissue model 2 was applied to the dynamic PIB-PET scans of 30 subjects (15 AD and 15 healthy controls). Parametric images of binding potential (BP_ND) and relative influx of tracer (R₁) were used to construct metric-specific DPs. Then Z-scores were assigned to the images based on its similarities with these DPs. Moreover, the subjects were classified based on a specific threshold derived from receiver-operator characteristic curves. Finally, a set of 32 different subjects diagnosed with AD (n = 4), Lewy body dementia (DLB) (n = 5), frontal temporal dementia (FTD) (n = 5), or mild cognitive impairment (MCI) with (+) (n = 11) or without (-) (n = 7) deposits of amyloid-β plaques, were tested against the DPs.

Results:Visual inspection of the DPs generated for each method were in line with previous studies: a higher binding (BP_ND images) and hypoperfusion (R₁) were observed in multiple gray matter regions of AD, as compared to healthy subjects. The Z-score threshold for classifying AD patients based on BP_ND and R₁ parametric maps were, respectively, 4.3 (area under the curve (AUC) = 1) and 1.3 (AUC = 0.9). Figure 1 shows the distribution of the Z-scores of the testing subjects. The AD group, as expected, presented an average value above the threshold, of 6.9 ± 6.8 (mean ± SD) for the BP_ND and 2.0 ± 1.7 for the R₁. Meanwhile, the MCI- and DLB groups both presented mean Z-scores smaller than the thresholds (for the DLB group, BP_ND was of 0.7 ± 1.0 and R₁ was 1.0 ± 0.8, and for the MCI-, 0.7 ± 3.1 and 0.8 ± 1.2 respectively). The MCI+ group presented a similar BP_ND pattern as the AD subjects, with a mean of 14.4 ± 7.2, but the same did not apply to the R₁, with a mean of 0.6 ± 0.6. Interestingly, an opposite effect was observed in the FTD group, with high R₁ Z-scores (2.1 ± 1.4), and low BP_ND (–0.4 ± 0.2).

Conclusions:
Pharmacokinetic modeling of dynamic PIB-PET scans provide high-quality parametric maps, such as BP_ND and R₁, that provide complimentary information. These images can be used as input for a SSM/PCA analysis, resulting in DPs that demonstrate different characteristic of AD patients when compared to healthy subjects. The multiparametric combination of these parametric images showed to be effective for a better discrimination of AD from other dementias.