Dicks, Ellen

Abstract There is a longstanding ambiguity regarding the clinical diagnosis of dementia syndromes predominantly targeting executive functions versus behaviour and personality. This is due to an incomplete understanding of the macro-scale anatomy underlying these symptomatologies, a partial overlap in clinical features and the fact that both phenotypes can emerge from the same pathology and vice versa. We collected data from a patient cohort of which 52 had dysexecutive Alzheimer’s disease, 30 had behavioural variant frontotemporal dementia (bvFTD), seven met clinical criteria for bvFTD but had Alzheimer’s disease pathology (behavioural Alzheimer’s disease) and 28 had amnestic Alzheimer’s disease. We first assessed group-wise differences in clinical and cognitive features and patterns of fluorodeoxyglucose (FDG) PET hypometabolism. We then performed a spectral decomposition of covariance between FDG-PET images to yield latent patterns of relative hypometabolism unbiased by diagnostic classification, which are referred to as ‘eigenbrains’. These eigenbrains were subsequently linked to clinical and cognitive data and meta-analytic topics from a large external database of neuroimaging studies reflecting a wide range of mental functions. Finally, we performed a data-driven exploratory linear discriminant analysis to perform eigenbrain-based multiclass diagnostic predictions. Dysexecutive Alzheimer’s disease and bvFTD patients were the youngest at symptom onset, followed by behavioural Alzheimer’s disease, then amnestic Alzheimer’s disease. Dysexecutive Alzheimer’s disease patients had worse cognitive performance on nearly all cognitive domains compared with other groups, except verbal fluency which was equally impaired in dysexecutive Alzheimer’s disease and bvFTD. Hypometabolism was observed in heteromodal cortices in dysexecutive Alzheimer’s disease, temporo-parietal areas in amnestic Alzheimer’s disease and frontotemporal areas in bvFTD and behavioural Alzheimer’s disease. The unbiased spectral decomposition analysis revealed that relative hypometabolism in heteromodal cortices was associated with worse dysexecutive symptomatology and a lower likelihood of presenting with behaviour/personality problems, whereas relative hypometabolism in frontotemporal areas was associated with a higher likelihood of presenting with behaviour/personality problems but did not correlate with most cognitive measures. The linear discriminant analysis yielded an accuracy of 82.1% in predicting diagnostic category and did not misclassify any dysexecutive Alzheimer’s disease patient for behavioural Alzheimer’s disease and vice versa. Our results strongly suggest a double dissociation in that distinct macro-scale underpinnings underlie predominant dysexecutive versus personality/behavioural symptomatology in dementia syndromes. This has important implications for the implementation of criteria to diagnose and distinguish these diseases and supports the use of data-driven techniques to inform the classification of neurodegenerative diseases.

Abstract Given the prevalence of dementia and the development of pathology-specific disease-modifying therapies, high-value biomarker strategies to inform medical decision-making are critical. In vivo tau-PET is an ideal target as a biomarker for Alzheimer’s disease diagnosis and treatment outcome measure. However, tau-PET is not currently widely accessible to patients compared to other neuroimaging methods. In this study, we present a convolutional neural network (CNN) model that imputes tau-PET images from more widely available cross-modality imaging inputs. Participants (n = 1192) with brain T1-weighted MRI (T1w), fluorodeoxyglucose (FDG)-PET, amyloid-PET and tau-PET were included. We found that a CNN model can impute tau-PET images with high accuracy, the highest being for the FDG-based model followed by amyloid-PET and T1w. In testing implications of artificial intelligence-imputed tau-PET, only the FDG-based model showed a significant improvement of performance in classifying tau positivity and diagnostic groups compared to the original input data, suggesting that application of the model could enhance the utility of the metabolic images. The interpretability experiment revealed that the FDG- and T1w-based models utilized the non-local input from physically remote regions of interest to estimate the tau-PET, but this was not the case for the Pittsburgh compound B-based model. This implies that the model can learn the distinct biological relationship between FDG-PET, T1w and tau-PET from the relationship between amyloid-PET and tau-PET. Our study suggests that extending neuroimaging’s use with artificial intelligence to predict protein specific pathologies has great potential to inform emerging care models.

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.

AbstractFrom a complex systems perspective, clinical syndromes emerging from neurodegenerative diseases are thought to result from multiscale interactions between aggregates of misfolded proteins and the disequilibrium of large-scale networks coordinating functional operations underpinning cognitive phenomena. Across all syndromic presentations of Alzheimer’s disease, age-related disruption of the default mode network is accelerated by amyloid deposition. Conversely, syndromic variability may reflect selective neurodegeneration of modular networks supporting specific cognitive abilities. In this study, we leveraged the breadth of the Human Connectome Project-Aging cohort of non-demented individuals (N = 724) as a normative cohort to assess the robustness of a biomarker of default mode network dysfunction in Alzheimer’s disease, the network failure quotient, across the aging spectrum. We then examined the capacity of the network failure quotient and focal markers of neurodegeneration to discriminate patients with amnestic (N = 8) or dysexecutive (N = 10) Alzheimer’s disease from the normative cohort at the patient level, as well as between Alzheimer’s disease phenotypes. Importantly, all participants and patients were scanned using the Human Connectome Project-Aging protocol, allowing for the acquisition of high-resolution structural imaging and longer resting-state connectivity acquisition time. Using a regression framework, we found that the network failure quotient related to age, global and focal cortical thickness, hippocampal volume, and cognition in the normative Human Connectome Project-Aging cohort, replicating previous results from the Mayo Clinic Study of Aging that used a different scanning protocol. Then, we used quantile curves and group-wise comparisons to show that the network failure quotient commonly distinguished both dysexecutive and amnestic Alzheimer’s disease patients from the normative cohort. In contrast, focal neurodegeneration markers were more phenotype-specific, where the neurodegeneration of parieto-frontal areas associated with dysexecutive Alzheimer’s disease, while the neurodegeneration of hippocampal and temporal areas associated with amnestic Alzheimer’s disease. Capitalizing on a large normative cohort and optimized imaging acquisition protocols, we highlight a biomarker of default mode network failure reflecting shared system-level pathophysiological mechanisms across aging and dysexecutive and amnestic Alzheimer’s disease and biomarkers of focal neurodegeneration reflecting distinct pathognomonic processes across the amnestic and dysexecutive Alzheimer’s disease phenotypes. These findings provide evidence that variability in inter-individual cognitive impairment in Alzheimer’s disease may relate to both modular network degeneration and default mode network disruption. These results provide important information to advance complex systems approaches to cognitive aging and degeneration, expand the armamentarium of biomarkers available to aid diagnosis, monitor progression and inform clinical trials.

Abstract Dysexecutive Alzheimer’s disease (dAD) manifests as a progressive dysexecutive syndrome without prominent behavioral features, and previous studies suggest clinico-radiological heterogeneity within this syndrome. We uncovered this heterogeneity using unsupervised machine learning in 52 dAD patients with multimodal imaging and cognitive data. A spectral decomposition of covariance between FDG-PET images yielded six latent factors (“eigenbrains”) accounting for 48% of variance in patterns of hypometabolism. These eigenbrains differentially related to age at onset, clinical severity, and cognitive performance. A hierarchical clustering on the eigenvalues of these eigenbrains yielded four dAD subtypes, i.e. “left-dominant,” “right-dominant,” “bi-parietal-dominant,” and “heteromodal-diffuse.” Patterns of FDG-PET hypometabolism overlapped with those of tau-PET distribution and MRI neurodegeneration for each subtype, whereas patterns of amyloid deposition were similar across subtypes. Subtypes differed in age at onset and clinical severity where the heteromodal-diffuse exhibited a worse clinical picture, and the bi-parietal had a milder clinical presentation. We propose a conceptual framework of executive components based on the clinico-radiological associations observed in dAD. We demonstrate that patients with dAD, despite sharing core clinical features, are diagnosed with variability in their clinical and neuroimaging profiles. Our findings support the use of data-driven approaches to delineate brain–behavior relationships relevant to clinical practice and disease physiology.

AbstractBackgroundGrey matter covariance networks become disrupted early in Alzheimer’s disease (AD), when amyloid starts aggregating, and before the onset of irreversible atrophy. However, the precise relationship of grey matter network breakdown and the AD pathophysiological process remains unclear. Here, we studied to what extent shared familial (genes and family environment) or environmental factors unique to a twin contribute to these relationships in a sample of older monozygotic twins with intact cognition.MethodFrom the EMIF‐AD PreclinAD study we included monozygotic twins who had at least an MRI scan (n=197) and CSF available (n=124). We assessed the upper limit of genetic contribution to grey matter network measures (size, degree, connectivity density, clustering, path length, small‐world measures) with within‐twin pair correlations. Associations between CSF AD markers (Aβ42/40, t‐tau, p‐tau) and other markers related to AD pathophysiology (neurogranin, Aβ38, Aβ40, BACE1) with grey matter network disruptions were assessed with linear mixed models. Twin analyses (cross‐twin cross‐trait, twin difference analyses) were used to assess the contribution of shared familial and/or unique environmental factors to these associations.ResultTwenty individuals (16%) had abnormal CSF Aβ42/40 levels. Grey matter network measures were highly correlated within twin pairs (r=0.54‐0.97; all p<0.001, Fig.1), suggesting a moderate to strong genetic background. Across the whole sample, more abnormal Aβ42/40 ratios were moderately related to lower connectivity density (β±SE = 0.176±0.09, p=0.055, Fig.2). The strongest associations with grey matter network measures were observed for t‐tau and p‐tau levels (range β=‐0.344‐0.182), followed by Aβ production markers (Aβ38, Aβ40, BACE1) (range β=‐0.202‐0.165). We observed no relationships between neurogranin and grey matter network measures. Cross‐twin cross‐trait analyses showed that levels for t‐tau, p‐tau and Aβ38 in one twin were associated with grey matter network measures in their co‐twin (r=‐0.27 to ‐0.21; p range=0.02 to 0.07, Fig.3). Within‐pair differences in BACE1 were related to within‐pair differences in normalized path length values (r=0.35; p=0.01; Fig.4).ConclusionOur results suggest a shared familial background for the relationship of grey matter network disruptions and AD‐related processes as measured in CSF. In addition, unique environmental factors influencing BACE1 may also affect grey matter network disruptions.

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.

AbstractBackgroundIndividuals with prodromal Alzheimer’s Disease (AD) and higher gray matter connectivity measures show slower decline, suggesting resilience to AD. However, the precise biological processes underlying preserved brain connections remains unclear. Here, we studied this question using cerebrospinal fluid (CSF) proteomics.MethodFrom the ADNI we selected 78 individuals with MCI and abnormal CSF amyloid (i.e., prodromal AD) and 34 with normal cognition and normal CSF amyloid levels for whom MRI and CSF proteomic data was available. For each T1 MRI scan we constructed gray matter networks using an automated pipeline (Tijms, Series et al. 2012) and calculated degree, connectivity density, clustering, path length and small‐world measures (i.e. gamma, lambda and small world coefficient). Linear models were used to test associations between each connectivity measure and CSF protein levels while correcting for age and gender. Proteins showing associations of p < 0.1 were selected to find groups of proteins that are associated with biological processes according to Gene Ontology (GO). Proteomic profiles for each connectivity measure were compared to investigate overlap. Finally, we tested for enrichment of potential upstream drivers with the Enrichr tool.ResultIn total, 115 proteins were associated with altered connectivity in prodromal AD subjects (figure 1). The majority of proteins (n = 93, 80%) showed higher levels with stronger (i.e., resilient) brain networks characterized by increased degree, connectivity density, clustering, path length, gamma, lambda or small world coefficient values. In controls, 80 of these 93 proteins showed an opposite relationship, with lower levels relating to higher values for these connectivity measures, which suggests that increased levels in prodromal AD may indicate a compensatory mechanism to maintain strong brain connectivity. According to a pathway analysis in GO, these proteins associated with stronger brain networks were mainly involved in neuroplasticity pathways. A subsequent Enrichr analysis suggested involvement of upstream divers SUZ12 and REST, which are known to regulate neuroplasticity processes.ConclusionWe observed that higher CSF levels of neuroplasticity associated proteins were associated with higher brain connectivity values in prodromal AD, suggesting that plasticity mechanisms may play a role in resilience to AD pathology.

AbstractBackgroundIn vivo tau‐positron emission tomography (PET) is an attractive biomarker for Alzheimer’s disease (AD) diagnosis and treatment. However, tau‐PET is less widely available than other modalities. In this study, we tested cross‐modality synthesis of tau‐PET brain images from fluorodeoxyglucose F‐18 (FDG)‐PET using a deep convolutional neural network (CNN).MethodParticipants (n=1,192) who had brain FDG‐PET with 18F‐FDG and tau‐PET with Flortaucipir (F‐18‐AV‐1451) were included for training and testing. This cohort spanned normal aging (ages 26‐98), pre‐clinical, and clinical AD and related disorders including the FTD and DLB spectrum. External validation was done using ADNI (n=288). The PET scans were co‐registered to the corresponding MRI and subsequently warped to Mayo Clinic Adult Lifespan Template (MCALT) space. Tau‐PET images were SUVR‐normalized to the cerebellar crus, and FDG to the pons. A 3D dense‐U‐net model was utilized as an architecture. Cross‐validation experiments were conducted using 5‐fold validations (60% training set, 20% validation set, and 20% test set) with mean squared error as the loss function.ResultOur dense‐U‐net model successfully synthesized tau‐PET from metabolic images with good correlation and low prediction error for regional SUVRs (Figure 1A‐C). The model showed a robust prediction ability, performing accurately in an independent, external ADNI cohort (Figure 1D‐F). The model‐imputed tau‐PET significantly improved performance in classifying tau positivity (mean AUROC(±SD)=0.78±0.04 and 0.85±0.03 for FDG‐PET and synthesized tau‐PET, respectively) and diagnostic groups (cognitively unimpaired with abnormal amyloid‐PET vs. cognitively impaired with abnormal amyloid‐PET) compared to the original input FDG data (mean AUROC(±SD)=0.89±0.04, 0.85±0.05 0.91±0.04 for actual tau‐PET, FDG‐PET and synthesized tau‐PET, respectively) (Figure 2), suggesting enhanced clinical utility for metabolic images. The ADNI cohort also showed similar results (for tau positivity: AUROC=0.66 and 0.78 for FDG‐PET and synthesized tau‐PET, respectively; for CU A+ vs. CI A+: AUROC=0.86, 0.62, and 0.73 for actual tau‐PET, FDG‐PET, and synthesized tau‐PET; Figure 3).ConclusionWe showed that using a CNN model to predict tau‐PET from FDG‐PET is feasible. The synthesized tau‐PET can augment the value of FDG‐PET, facilitating the multi‐modal diagnosis of AD.

AbstractBackgroundGrey matter covariance networks become disrupted early in Alzheimer’s disease (AD), when amyloid starts aggregating, and before the onset of irreversible atrophy. However, the precise relationship of grey matter network breakdown and the AD pathophysiological process remains unclear. Here, we studied to what extent shared familial (genes and family environment) or environmental factors unique to a twin contribute to these relationships in a sample of older monozygotic twins with intact cognition.MethodFrom the EMIF‐AD PreclinAD study we included monozygotic twins who had at least an MRI scan (n=197) and CSF available (n=124). We assessed the upper limit of genetic contribution to grey matter network measures (size, degree, connectivity density, clustering, path length, small‐world measures) with within‐twin pair correlations. Associations between CSF AD markers (Aβ42/40, t‐tau, p‐tau) and other markers related to AD pathophysiology (neurogranin, Aβ38, Aβ40, BACE1) with grey matter network disruptions were assessed with linear mixed models. Twin analyses (cross‐twin cross‐trait, twin difference analyses) were used to assess the contribution of shared familial and/or unique environmental factors to these associations.ResultTwenty individuals (16%) had abnormal CSF Aβ42/40 levels. Grey matter network measures were highly correlated within twin pairs (r=0.54‐0.97; all p<0.001, Fig.1), suggesting a moderate to strong genetic background. Across the whole sample, more abnormal Aβ42/40 ratios were moderately related to lower connectivity density (β±SE = 0.176±0.09, p=0.055, Fig.2). The strongest associations with grey matter network measures were observed for t‐tau and p‐tau levels (range β=‐0.344‐0.182), followed by Aβ production markers (Aβ38, Aβ40, BACE1) (range β=‐0.202‐0.165). We observed no relationships between neurogranin and grey matter network measures. Cross‐twin cross‐trait analyses showed that levels for t‐tau, p‐tau and Aβ38 in one twin were associated with grey matter network measures in their co‐twin (r=‐0.27 to ‐0.21; p range=0.02 to 0.07, Fig.3). Within‐pair differences in BACE1 were related to within‐pair differences in normalized path length values (r=0.35; p=0.01; Fig.4).ConclusionOur results suggest a shared familial background for the relationship of grey matter network disruptions and AD‐related processes as measured in CSF. In addition, unique environmental factors influencing BACE1 may also affect grey matter network disruptions.