Lei Wang - US grants

Project Summary Over the past decade, we have been building, parsing and wrangling systems for extracting neurodegeneration and neurodevelopment biomarkers from high-dimensional magnetic resonance (MR) imagery at 1 mm3 scale which are discriminating. At the same time, large and complex data sets and networks of segmented structures are becoming increasingly available to the research community such as Predict-HD, Track-HD, ADNI, and SchizConnect. Neuroscientists and clinicians are interested in tracking biomarkers which characterize rates of atrophy in anatomical networks, onset of or changepoint times of spread through the networks, and prediction of risk to conversion as determined by clinical symptoms. These wrangling and modeling methods are novel. Our biomarkers are extracted via brain mapping technologies based on diffeomorphometry, the study of morphological change via diffeomorphic tracking of anatomical coordinate systems at the sub millimeter scale. Like stereology, diffeomorphometry discovers high-dimensional features signalling neurodegeneration and neurodevelopment via tight integration of random field based statistical methods via large deviation empirical probability estimators calculated via high-dimensional permutation testing. Family-wise rates are calculated for group comparisons, and have been advanced changepoint modelling allowing us to explicitly estimate the spread of progression of anatomical feature change through the networked structures associated to neurodegeneration - Alzheimer's Disease (AD) and Huntingdon's Disease (HD) and neurodevelopment - Schizophrenia (SZ) and Attention Deficit and Hyperactive Disorder (ADHD). These tools will be disseminated and tested via MriCloud. We will perform three specific aims. Aim 1 will use our MriCloud architecture to deploy a Multi-Atlas Brain Mapping module for mapping an ontology of approximately 400 structures to T1 and DTI data. The architecture will support many atlases which are matched across a broad range of age from pediatric to geriatric groups, and as well as several diseases. Aim 2 will deploy a Statistical Shape Diffeomorphometry module consisting of pipelines for a) generating templates of structures from populations of cross-sectional datasets, b) data reduction to templates for cross-sectional and longitudinal geodesic mappings, and c) multiple hypothesis testing procedures based on vertex, Laplace-Beltrami basis functions and PCA basis functions. Users with their own ontology definitions of the subcortical structures will be able to generate population templates and visualize the statistics in template coordinates. Aim 3 will generate a webportal for users to use modules from Aims 1 and 2 to examine abnormalities in networks of structures such as the striatum, thalamus, amygdala and hippocampus.

PROJECT SUMMARY Alzheimer's dementia (AD) is the most common form of dementia in adults over the age of 65, and Frontotemporal dementia (FTD) is the leading cause of dementia in middle age, with the behavioral variant subtype (bvFTD) being the most prevalent form. The relationships between clinical syndromes and pathological causes are complex, which makes accurate diagnosis difficult. For example, multiple studies have indicated that a significant proportion of cases of AD-like dementia show evidence of non-AD pathology, such as inclusions of the transactive response DNA-binding protein 43 (TDP-43), a protein associated with clinical FTD. Also, AD neuropathology has been found in 15?30% of patient with the clinical diagnosis of frontotemporal dementia (FTD). As treatment agents with potential disease-modifying effects are developed, sensitive and specific biomarkers will be needed, so that they can be tested and then eventually used in the appropriate patient populations. In this project, we will focus on clinically diagnosed bvFTD and AD patients, and apply machine learning to multimodal neuroimaging (T1, FDG-PET) data pooled from large, multisite studies of AD and FTD. Our goal is to develop novel biomarkers that can differentiate bvFTD, AD and controls. Our hypothesis is that each neuropathology is associated with a distinct biomarker signature, and these signatures can be discovered through well-characterized clinical, neurological and neuroanatomical profiles. We will use available amyloid imaging and cerebrospinal fluid (CSF) measures of ?-amyloid and tau to assess the robustness of our predictions of AD neuropathologies. In Aim 1 we will use cross-sectional and longitudinal structural imaging to develop predictive biomarker models for differentiating bvFTD vs. AD. In Aim 2 we will use cross-sectional and longitudinal FDG-PET imaging to develop predictive biomarker models. In Aim 3 we will evaluate the combination of structural and FDG-PET imaging as predictive biomarker models. Relevance: This research supports NIH initiatives on long-term, personalized precision medicine and big data science. Our predictive biomarker models can inform participant selection in clinical trials so that we can identify disease-modifying treatments with greater power. Our system-biology approach can enable us to generate new questions on mechanisms underlying the origin and progression of neuro-pathological processes, create new data and computational tools that can in turn generate new insights and new hypotheses.

In replication, mega analysis, and meta-analysis that are critical to the advancement of neuroimaging research, a critical question is how to harness already-collected data for replication purposes efficiently and rigorously. Much of the present efforts on reproducibility science assumes that appropriate datasets are available. While many different neuroimaging databases exist, they have different languages, formats, and usually do not communicate with each other. Moreover, neuroimaging data are collected in hundreds of laboratories each year, forming the ?long tail of science? data. Much of this data are described in journal publications but remain underutilized. A critical gap therefore exists in getting enough data of the right kind to the scientist. We propose the NeuroBridge, a data sharing infrastructure that combines innovative techniques and established technologies in a platform that will greatly enhance the reuse of critical neuroscience data sets. RELEVANCE (See instructions): Our research develops technologies that address critical gaps in big data clinical neuroscience research. Our proposed research includes collaboration with addiction projects such as COINSTAC and ENIGMA- addiction. These efforts will leverage technologies that can ?expand access to increasingly larger databases ? along with rapid advances in analytic, computational, and information technologies,? and increase scientific rigor and reproducibility that will lead to ?improved quality and credibility of addiction research and advancement in research on drug use and addiction,? key cross-cutting themes of the National Institute on Drug Abuse (NIDA). Moreover, our focus on developing advanced informatics tools for ?Big Data and Computational Science? is in direct alignment with the goals of the Division of Neuroscience and Behavior (DNB) with NIDA. Our above proposed collaborations can further the NIDA mission of ?identifying the neural circuits underlying drug addiction and their functional properties,? and ?the causes and consequences of drugs of abuse and addiction across the lifespan and to guide treatment strategies.?