Dicks, Ellen

AbstractBackgroundFluorodeoxyglucose positron emission tomography (FDG‐PET) is an established modality for differential diagnosis of dementia. Deriving phenotypic signatures from FDG‐PET via machine learning is challenging due to the high dimensional nature of brain images relative to the generally small number of labeled examples available for training, the class imbalance among those labels, and the cooccurrence of multiple pathologies. In this study, we developed a multi‐class, multi‐label framework to address these challenges.MethodA database of clinically acquired PET/CT images from 3,000 unique patients was used to develop a latent space model using matrix decomposition. This model was then applied to images from a separate cohort of Mayo Clinic Alzheimer’s Disease Research Center participants (n=1,745) labeled as cognitively unimpaired (CU) (n=1,436) or with the following potentially co‐occurring phenotypes: Alzheimer’s disease (AD) (n=165), Lewy body dementia (DLB) (n=92), behavioral variant frontotemporal dementia (bvFTD) (n=43), semantic (svPPA) (n=10) and logopenic (lvPPA) (n=13) variant PPA, and posterior cortical atrophy (PCA) (n=17). A k‐nearest neighbors classifier that is robust to these imbalanced and overlapping labels was then trained on these examples. The resulting classifier was evaluated by area under receiver‐operator characteristic curve (ROC‐AUC) via leave one out cross validation, using clinical diagnosis as the gold standard.ResultROC curves and AUC scores for each phenotype are illustrated in Fig. 1a. Because the classifier is based on a k‐nearest neighbors connectivity matrix, it has a convenient graphical representation, where images are nodes and edges are drawn between an image and its set of nearest neighbors in latent space. A self‐organizing force directed graph constructed in this way is illustrated in Fig 1b, highlighting the strong separation of CU and degenerative images, as well as the segregation of each phenotype within the neurodegenerative region of the graph.ConclusionIn this study, we developed a machine learning framework for classification of neurodegenerative disease based on k‐nearest neighbor analysis in a low dimensional latent space projection of FDG‐PET images. By leveraging low‐dimensional representations and k‐nearest neighbors analysis, this framework is robust in multi‐class, multi‐label tasks with strong class imbalance and provides a highly interpretable graphical representation.

AbstractBackgroundBrain amyloid‐β (Aβ) is the pathological hallmark of Alzheimer's disease (AD). In logistic disease models, Aβ accumulation is a sigmoid function of time‐since‐disease‐onset (TSDO) (figure 1). Previous positron emission tomography (PET)‐based models vary accumulation onset(t50) and duration(r) globally; capacity(K) and baseline(NS) regionally (Whittington2018). We confirm existing approaches and propose a more powerful ICA‐based approach to quantify disease severity and estimate TSDO.MethodWe used 1071 18F‐florbetapir standard uptake value ratio (SUVR) images from the ADNI‐2 study (adni.loni.usc.edu/data‐samples/data‐types/pet). Images were mapped into MNI space. Averages were extracted using the Harvard‐Oxford brain‐atlas. Whole‐brain tracer‐specific sigmoid parameters (Jack2013) obtained from the literature were used to estimate TSDO.Of 16 models of regional Aβ accumulation (each of the 4 regional sigmoid parameters varied either regionally or globally), the optimal Bayesian information criterion was found with global t50 and r, and regional NS and K (figure 1) with global values r=6.16y and t50=4.10y.Linearised maps of NS and K were obtained by regressing the SUVR maps onto the global sigmoid. We also estimated these maps as independent components, using a 2‐component ICA on the SUVR maps. Both outcomes were used to quantify Aβ accumulation from SUVR images as weighting factors of the accumulation map.We compared the weights from the logistic model and the ICA model in ADNI, using effect size measured with Hedges' g between cognitively normal (CN), subjective memory complaints (SMC), mild cognitive impairment (EMCI/MCI/LMCI) and AD groups. We compared 3 longitudinal visits (N=112) in the OASIS‐3 study (see www.oasis‐brains.org) with both methods, global SUVR and Centiloid (Klunk2015) using 11C‐PiB PET SUVR images.ResultMaps of accumulation capacity from both models had spatial correlation of 0.86 (figure 2); baseline maps had spatial correlation of 0.95. Hedges' g between ADNI groups was 2.25 for K, and 2.42 for ICA (1.46 for global SUVR). In OASIS‐3, Hedges' g between visits was 1.24 for K, 1.46 for ICA (global SUVR 0.15, Centiloid 0.4).ConclusionWe demonstrate that linear accumulation models can be used to quantify brain Aβ with PET; maps obtained by ICA yield larger effect sizes than the logistic method for differentiating groups and measuring changes between visits.

AbstractIndividuals with prodromal Alzheimer’s disease show considerable variability in rates of cognitive decline, which hampers the ability to detect potential treatment effects in clinical trials. Prognostic markers to select those individuals who will decline rapidly within a trial time frame are needed. Brain network measures based on grey matter covariance patterns have been associated with future cognitive decline in Alzheimer’s disease. In this longitudinal cohort study, we investigated whether cut-offs for grey matter networks could be derived to detect fast disease progression at an individual level. We further tested whether detection was improved by adding other biomarkers known to be associated with future cognitive decline [i.e. CSF tau phosphorylated at threonine 181 (p-tau181) levels and hippocampal volume]. We selected individuals with mild cognitive impairment and abnormal CSF amyloid β1–42 levels from the Amsterdam Dementia Cohort and the Alzheimer’s Disease Neuroimaging Initiative, when they had available baseline structural MRI and clinical follow-up. The outcome was progression to dementia within 2 years. We determined prognostic cut-offs for grey matter network properties (gamma, lambda and small-world coefficient) using time-dependent receiver operating characteristic analysis in the Amsterdam Dementia Cohort. We tested the generalization of cut-offs in the Alzheimer’s Disease Neuroimaging Initiative, using logistic regression analysis and classification statistics. We further tested whether combining these with CSF p-tau181 and hippocampal volume improved the detection of fast decliners. We observed that within 2 years, 24.6% (Amsterdam Dementia Cohort, n = 244) and 34.0% (Alzheimer’s Disease Neuroimaging Initiative, n = 247) of prodromal Alzheimer’s disease patients progressed to dementia. Using the grey matter network cut-offs for progression, we could detect fast progressors with 65% accuracy in the Alzheimer’s Disease Neuroimaging Initiative. Combining grey matter network measures with CSF p-tau and hippocampal volume resulted in the best model fit for classification of rapid decliners, increasing detecting accuracy to 72%. These data suggest that single-subject grey matter connectivity networks indicative of a more random network organization can contribute to identifying prodromal Alzheimer’s disease individuals who will show rapid disease progression. Moreover, we found that combined with p-tau and hippocampal volume this resulted in the highest accuracy. This could facilitate clinical trials by increasing chances to detect effects on clinical outcome measures.

Abstract Structural grey matter covariance networks provide an individual quantification of morphological patterns in the brain. The network integrity is disrupted in sporadic Alzheimer’s disease, and network properties show associations with the level of amyloid pathology and cognitive decline. Therefore, these network properties might be disease progression markers. However, it remains unclear when and how grey matter network integrity changes with disease progression. We investigated these questions in autosomal dominant Alzheimer’s disease mutation carriers, whose conserved age at dementia onset allows individual staging based upon their estimated years to symptom onset. From the Dominantly Inherited Alzheimer Network observational cohort, we selected T1-weighted MRI scans from 269 mutation carriers and 170 non-carriers (mean age 38 ± 15 years, mean estimated years to symptom onset −9 ± 11), of whom 237 had longitudinal scans with a mean follow-up of 3.0 years. Single-subject grey matter networks were extracted, and we calculated for each individual the network properties which describe the network topology, including the size, clustering, path length and small worldness. We determined at which time point mutation carriers and non-carriers diverged for global and regional grey matter network metrics, both cross-sectionally and for rate of change over time. Based on cross-sectional data, the earliest difference was observed in normalized path length, which was decreased for mutation carriers in the precuneus area at 13 years and on a global level 12 years before estimated symptom onset. Based on longitudinal data, we found the earliest difference between groups on a global level 6 years before symptom onset, with a greater rate of decline of network size for mutation carriers. We further compared grey matter network small worldness with established biomarkers for Alzheimer disease (i.e. amyloid accumulation, cortical thickness, brain metabolism and cognitive function). We found that greater amyloid accumulation at baseline was associated with faster decline of small worldness over time, and decline in grey matter network measures over time was accompanied by decline in brain metabolism, cortical thinning and cognitive decline. In summary, network measures decline in autosomal dominant Alzheimer’s disease, which is alike sporadic Alzheimer’s disease, and the properties show decline over time prior to estimated symptom onset. These data suggest that single-subject networks properties obtained from structural MRI scans form an additional non-invasive tool for understanding the substrate of cognitive decline and measuring progression from preclinical to severe clinical stages of Alzheimer’s disease.