Optimal Target Region for Subject Classification on the Basis of Amyloid PET Images

Abstract

Classification of subjects on the basis of amyloid PET scans is increasingly being used in research studies and clinical practice. Although qualitative, visual assessment is currently the gold standard approach, automated classification techniques are inherently more reproducible and efficient. The objective of this work was to develop a statistical approach for the automated classification of subjects with different levels of cognitive impairment into a group with low amyloid levels (Aβ L ) and a group with high amyloid levels (Aβ H ) through the use of amyloid PET data from the Alzheimer Disease Neuroimaging Initiative study. Methods: In our framework, an iterative, voxelwise, regularized discriminant analysis is combined with a receiver operating characteristic approach that optimizes the selection of a region of interest (ROI) and a cutoff value for the automated classification of subjects into the Aβ L and Aβ H groups. The robustness, spatial stability, and generalization of the resulting target ROIs were evaluated by use of the standardized uptake value ratio for 18 F-florbetapir PET images from subjects who served as healthy controls, subjects who had mild cognitive impairment, and subjects who had Alzheimer disease and were participating in the Alzheimer Disease Neuroimaging Initiative study. Results: We determined that several iterations of the discriminant analysis improved the classification of subjects into the Aβ L and Aβ H groups. We found that an ROI consisting of the posterior cingulate cortex/precuneus and the medial frontal cortex yielded optimal group separation and showed good stability across different reference regions and cognitive cohorts. A key step in this process was the automated determination of the cutoff value for group separation, which was dependent on the reference region used for the standardized uptake value ratio calculation and which was shown to have a relatively narrow range across subject groups. Conclusion: We developed a data-driven approach for the determination of an optimal target ROI and an associated cutoff value for the separation of subjects into the Aβ L and Aβ H groups. Future work should include the application of this process to other datasets to facilitate the determination of the translatability of the optimal ROI obtained in this study to other populations. Ideally, the accuracy of our target ROI and cutoff value could be further validated with PET-autopsy data from large-scale studies. It is anticipated that this approach will be extremely useful for the enrichment of study populations in clinical trials involving putative disease-modifying therapeutic agents for Alzheimer disease. The advent of PET radiotracers with high specificity for b-amyloid plaques represents a paradigm shift in Alzheimer disease (AD) research. The use of amyloid PET has markedly improved the understanding of the relationship between b-amyloid pathology and cognition (1-3), brain structure (4,5), cerebral glucose metabolism (6-8), and brain connectivity (9-11). In addition, amyloid...

Optimizing PiB-PET SUVR change-over-time measurement by a large-scale analysis of longitudinal reliability, plausibility, separability, and correlation with MMSE

“…We also leave for future work a more exhaustive comparison of PVC methods with more variations, e.g. ROI-based methods (Rousset et al, 1998) and comparisons with methods using data-driven selections of individual voxels (Borghammer et al, 2009; Carbonell et al, 2015; Razifar et al, 2009). We also did not examine PET-only methods (Bilgel et al, 2015; Fripp et al, 2008; Raniga et al, 2007; Zhou et al, 2014) because MRI is generally available in most studies that include amyloid PET.…”

Contributions of imprecision inPET‐MRIrigid registration to imprecision in amyloidPETSUVRmeasurements

“…In longitudinal studies measuring change over time, precision of serial measurements is particularly important [Schmidt et al, ]. Prior works have primarily focused upon other relevant parameters such as use of PVC, choice of target voxels, and choice of reference region [Brendel et al, ; Carbonell et al, ; Chen et al, ; Landau et al, ; Schwarz et al, ]. In this work, we focus specifically on a relatively under‐studied measurement step: cross‐modality rigid registration.…”

Modulation of glucose metabolism and metabolic connectivity by β-amyloid

“…The target ROI for SUVR measurement (SUVR ROI ) for the present work was the statistically derived ROI from our previous analysis of [18F]-florbetapir images from the same subjects. 31 Briefly, this iterative process is initiated by labeling subjects into low (Ab L ) and high (Ab H ) b-amyloid levels based on the global average SUVR. A voxelwise, regularized discriminant analysis (RDA) is then employed to identify an optimal target region for classification and receiver operating characteristic (ROC) analysis in order to identify the optimal threshold for classification.…”