Thompson, Paul M.

Background: Disease progression prediction based on neuroimaging biomarkers is vital in Alzheimer’s disease (AD) research. Convolutional neural networks (CNN) have been proved to be powerful for various computer vision research by refining reliable and high-level feature maps from image patches. Objective: A key challenge in applying CNN to neuroimaging research is the limited labeled samples with high dimensional features. Another challenge is how to improve the prediction accuracy by joint analysis of multiple data sources (i.e., multiple time points or multiple biomarkers). To address these two challenges, we propose a novel multi-task learning framework based on CNN. Methods: First, we pre-trained CNN on the ImageNet dataset and transferred the knowledge from the pre-trained model to neuroimaging representation. We used this deep model as feature extractor to generate high-level feature maps of different tasks. Then a novel unsupervised learning method, termed Multi-task Stochastic Coordinate Coding (MSCC), was proposed for learning sparse features of multi-task feature maps by using shared and individual dictionaries. Finally, Lasso regression was performed on these multi-task sparse features to predict AD progression measured by the Mini-Mental State Examination (MMSE) and the Alzheimer’s Disease Assessment Scale cognitive subscale (ADAS-Cog). Results: We applied this novel CNN-MSCC system on the Alzheimer’s Disease Neuroimaging Initiative dataset to predict future MMSE/ADAS-Cog scales. We found our method achieved superior performances compared with seven other methods. Conclusion: Our work may add new insights into data augmentation and multi-task deep model research and facilitate the adoption of deep models in neuroimaging research.

INTRODUCTION The cerebral cortex underlies our complex cognitive capabilities. Variations in human cortical surface area and thickness are associated with neurological, psychological, and behavioral traits and can be measured in vivo by magnetic resonance imaging (MRI). Studies in model organisms have identified genes that influence cortical structure, but little is known about common genetic variants that affect human cortical structure. RATIONALE To identify genetic variants associated with human cortical structure at both global and regional levels, we conducted a genome-wide association meta-analysis of brain MRI data from 51,665 individuals across 60 cohorts. We analyzed the surface area and average thickness of the whole cortex and 34 cortical regions with known functional specializations. RESULTS We identified 369 nominally genome-wide significant loci ( P < 5 × 10 −8 ) associated with cortical structure in a discovery sample of 33,992 participants of European ancestry. Of the 360 loci for which replication data were available, 241 loci influencing surface area and 66 influencing thickness remained significant after replication, with 237 loci passing multiple testing correction ( P < 8.3 × 10 −10 ; 187 influencing surface area and 50 influencing thickness). Common genetic variants explained 34% (SE = 3%) of the variation in total surface area and 26% (SE = 2%) in average thickness; surface area and thickness showed a negative genetic correlation ( r G = −0.32, SE = 0.05, P = 6.5 × 10 −12 ), which suggests that genetic influences have opposing effects on surface area and thickness. Bioinformatic analyses showed that total surface area is influenced by genetic variants that alter gene regulatory activity in neural progenitor cells during fetal development. By contrast, average thickness is influenced by active regulatory elements in adult brain samples, which may reflect processes that occur after mid-fetal development, such as myelination, branching, or pruning. When considered together, these results support the radial unit hypothesis that different developmental mechanisms promote surface area expansion and increases in thickness. To identify specific genetic influences on individual cortical regions, we controlled for global measures (total surface area or average thickness) in the regional analyses. After multiple testing correction, we identified 175 loci that influence regional surface area and 46 that influence regional thickness. Loci that affect regional surface area cluster near genes involved in the Wnt signaling pathway, which is known to influence areal identity. We observed significant positive genetic correlations and evidence of bidirectional causation of total surface area with both general cognitive functioning and educational attainment. We found additional positive genetic correlations between total surface area and Parkinson’s disease but did not find evidence of causation. Negative genetic correlations were evident between total surface area and insomnia, attention deficit hyperactivity disorder, depressive symptoms, major depressive disorder, and neuroticism. CONCLUSION This large-scale collaborative work enhances our understanding of the genetic architecture of the human cerebral cortex and its regional patterning. The highly polygenic architecture of the cortex suggests that distinct genes are involved in the development of specific cortical areas. Moreover, we find evidence that brain structure is a key phenotype along the causal pathway that leads from genetic variation to differences in general cognitive function. Identifying genetic influences on human cortical structure. ( A ) Measurement of cortical surface area and thickness from MRI. ( B ) Genomic locations of common genetic variants that influence global and regional cortical structure. ( C ) Our results support the radial unit hypothesis that the expansion of cortical surface area is driven by proliferating neural progenitor cells. ( D ) Cortical surface area shows genetic correlation with psychiatric and cognitive traits. Error bars indicate SE. IMAGE CREDITS: (A) K. COURTNEY; (C) M. R. GLASS

Abstract Introduction The overall goal of the Alzheimer's Disease Neuroimaging Initiative (ADNI) is to validate biomarkers for Alzheimer's disease (AD) clinical trials. ADNI is a multisite, longitudinal, observational study that has collected many biomarkers since 2004. Recent publications highlight the multifactorial nature of late‐onset AD. We discuss selected topics that provide insights into AD progression and outline how this knowledge may improve clinical trials. Methods We used standard methods to identify nearly 600 publications using ADNI data from 2016 and 2017 (listed in Supplementary Material and searchable at http://adni.loni.usc.edu/news‐publications/publications/ ). Results (1) Data‐driven AD progression models supported multifactorial interactions rather than a linear cascade of events. (2) β‐Amyloid (Aβ) deposition occurred concurrently with functional connectivity changes within the default mode network in preclinical subjects and was followed by specific and progressive disconnection of functional and anatomical networks. (3) Changes in functional connectivity, volumetric measures, regional hypometabolism, and cognition were detectable at subthreshold levels of Aβ deposition. 4. Tau positron emission tomography imaging studies detailed a specific temporal and spatial pattern of tau pathology dependent on prior Aβ deposition, and related to subsequent cognitive decline. 5. Clustering studies using a wide range of modalities consistently identified a “typical AD” subgroup and a second subgroup characterized by executive impairment and widespread cortical atrophy in preclinical and prodromal subjects. 6. Vascular pathology burden may act through both Aβ dependent and independent mechanisms to exacerbate AD progression. 7. The APOE ε4 allele interacted with cerebrovascular disease to impede Aβ clearance mechanisms. 8. Genetic approaches identified novel genetic risk factors involving a wide range of processes, and demonstrated shared genetic risk for AD and vascular disorders, as well as the temporal and regional pathological associations of established AD risk alleles. 9. Knowledge of early pathological changes guided the development of novel prognostic biomarkers for preclinical subjects. 10. Placebo populations of randomized controlled clinical trials had highly variable trajectories of cognitive change, underscoring the importance of subject selection and monitoring. 11. Selection criteria based on Aβ positivity, hippocampal volume, baseline cognitive/functional measures, and APOE ε4 status in combination with improved cognitive outcome measures were projected to decrease clinical trial duration and cost. 12. Multiple concurrent therapies targeting vascular health and other AD pathology in addition to Aβ may be more effective than single therapies. Discussion ADNI publications from 2016 and 2017 supported the idea of AD as a multifactorial disease and provided insights into the complexities of AD disease progression. These findings guided the development of novel biomarkers and suggested that subject selection on the basis of multiple factors may lower AD clinical trial costs and duration. The use of multiple concurrent therapies in these trials may prove more effective in reversing AD disease progression.

AbstractIntroduction18F‐fluorodeoxyglucose (FDG) positron emission tomography (PET) is commonly used to estimate neuronal injury in Alzheimer's disease (AD). Here, we evaluate the utility of dynamic PET measures of perfusion using 11C‐Pittsburgh compound B (PiB) to estimate neuronal injury in comparison to FDG PET.MethodsFDG, early frames of PiB images, and relative PiB delivery rate constants (PiB‐R1) were obtained from 110 participants from the Dominantly Inherited Alzheimer Network. Voxelwise, regional cross‐sectional, and longitudinal analyses were done to evaluate the correlation between images and estimate the relationship of the imaging biomarkers with estimated time to disease progression based on family history.ResultsMetabolism and perfusion images were spatially correlated. Regional PiB‐R1 values and FDG, but not early frames of PiB images, significantly decreased in the mutation carriers with estimated year to onset and with increasing dementia severity.DiscussionHypometabolism estimated by PiB‐R1 may provide a measure of brain perfusion without increasing radiation exposure.

T1-weighted MRI has been extensively used to extract imaging biomarkers and build classification models for differentiating Alzheimer’s disease (AD) patients from healthy controls, but only recently have brain connectome networks derived from diffusion-weighted MRI been used to model AD progression and various stages of disease such as mild cognitive impairment (MCI). MCI, as a possible prodromal stage of AD, has gained intense interest recently, since it may be used to assess risk factors for AD. Little work has been done to combine information from both white matter and gray matter, and it is unknown how much classification power the diffusion-weighted MRI-derived structural connectome could provide beyond information available from T1-weighted MRI. In this paper, we focused on investigating whether diffusion-weighted MRI-derived structural connectome can improve differentiating healthy controls subjects from those with MCI. Specifically, we proposed a novel feature-ranking method to build classification models using the most highly ranked feature variables to classify MCI with healthy controls. We verified our method on two independent cohorts including the second stage of Alzheimer’s Disease Neuroimaging Initiative (ADNI2) database and the National Alzheimer’s Coordinating Center (NACC) database. Our results indicated that 1) diffusion-weighted MRI-derived structural connectome can complement T1-weighted MRI in the classification task; 2) the feature-rank method is effective because of the identified consistent T1-weighted MRI and network feature variables on ADNI2 and NACC. Furthermore, by comparing the top-ranked feature variables from ADNI2, NACC, and combined dataset, we concluded that cross-validation using independent cohorts is necessary and highly recommended.