Bruce Fischl - US grants

DESCRIPTION (provided by applicant): Neurodegenerative and psychiatric disorders, as well as healthy aging, are all frequently associated with structural changes in the brain. These changes can cause alterations in the imaging properties of brain tissue, as well as changes in morphometric properties of the brain, such as volume, folding and surface area. This can be problematic, as analysis techniques that quantify morphometric changes without accounting for variability in the imaging properties of the tissue are liable to generate erroneous results in situations in which the tissue parameters have changed. In this grant application support is sought to construct a set of accurate and automated tools for the analysis of structural neuroimaging data. These tools will quantify alterations in brain morphometry, as well as changes in the tissue parameters that give rise to image contrast in magnetic resonance images. It is hypothesized that explicitly basing the analysis tools upon knowledge of the underlying physical principles that govern the imaging process will allow the characterization of subtle structural changes that have previously gone undetected. Aim 1 of this application is to develop a set of scans and optimization techniques that will allow for the accurate estimation of the underlying tissue parameters (i.e., T1, T2, proton density). Additional effort will be focused on removing various sources of inaccuracies in the data acquisition. This includes using optimization techniques to derive MR protocols with optimal contrast-to-noise, as well as the correction of various sources of distortion that arise in MR images such as gradient nonlinearities and real-time online correction of within-scan subject motion. This latter technique is of particular importance, as it will allow the tools to be applied to patient populations for which within-scan motion is frequently problematic. Aim 2 is to use a database of manually labeled datasets as the basis for the construction of an automated whole-brain segmentation procedure designed to assign a neuroanatomical label to every voxel in the brain (e.g. thalamus, caudate, putamen, etc). The segmentation procedure will disambiguate structures with similar tissue properties based on their location within the brain, as well as their spatial relationships to neighboring structures, encoded using an anisotropic markov random field. It is important to note that basing the segmentation upon the intrinsic tissue parameters renders the procedure largely insensitive to the details of a particular pulse sequence. Aim 3 is to employ the multi-spectral tissue parameters as the basis for surface-based morphometric analysis. The final aim is to validate the accuracy of the procedures, as well as their robustness to changes in scanner protocol. Upon completion of tool development they will be applied to the study of a variety of disorders, focusing on schizophrenia. Alzheimer's and Huntington's disease.

DESCRIPTION (provided by applicant): This proposal requests funding for a Shared Instrument to support a new, state of the art, high performance computational cluster for structural and functional image analysis. Specifically, we are proposing a 344 node cluster, comprised of 2.8 Ghz Xenon processors, that will greatly enhance the ability of users of the Martinos Center to conduct their currently funded research projects. Four broad areas of users are identified, and will realize a dramatic benefit: a) fMRI studies (to increase the throughput of surface analysis and statistical characterization), b) voxel-based morphometry studies (to increase throughput on large morphometry studies), c) multimodal integration studies (to facilitate complex forward and inverse modeling used in EEGIMEG/MRI and diffuse optical tomography) and d) algorithm development users (to enhance turnaround on technique optimization, validation and implementation.) A common feature that makes this migration possible is the use of the Freesurfer processing stream. This software already has in place a comprehensive, demonstrated infrastructure for distributed processing. Each of these classes of users has operational solutions now that utilize an existing but out-dated multi-mode compute server. Each of these users, however, will benefit from a more advanced server and the increased capabilities it engenders. The user community for this proposed instrument is broad; spanning 4 institutions (MGH, BWH, MIT, and BU) and many departments within these institutions. Also, this instrument enhances the performances capabilities of two Regional Resources, enabling these facilities to deliver greater computational power to their user.

[unreadable] DESCRIPTION (provided by applicant): By fusing information from multiple imaging modalities, multimodal imaging can provide substantially greater biological information content and detection power than the individual modalities employed in isolation. Despite the promise of the multimodal imaging framework, integration of multiple imaging modalities is often hindered in practice by the need to integrate advanced software applications which were not designed for interoperability. For example, integration of separate analysis environments requires reconciling different coordinate systems, harmonizing quality assurance goals, and integrating developer knowledge. While the unique challenges of software engineering for multimodal imaging are recognized in the field, no specific software methodology has been developed yet to address these challenges. In this grant, we propose to develop a set of software engineering methods for multimodality brain imaging. The project consists of two software engineering aims, and a driving biological aim: (1) integrate and modernize an anatomical analysis package (Free Surfer) and a diffusion tensor analysis package (Free Diffusion). (2) develop software methods for automatic failure detection in multimodal imaging; and (3) apply multimodal framework to integrate high resolution structural MRI and high angular resolution diffusion MRI. The multimodal structural-diffusion MRI framework will be employed to investigate the relationship between gray matter cortical thickness and the white matter microstructural integrity in the aging brain. This approach will allow us to investigate the correlation between the cortical structure and long-range connectional architecture of the aging brain, a key question in the physiology of aging. Development of a robust software engineering framework for integrating multiple image analysis environments promises to significantly accelerate the development of multimodal brain imaging methods and allow for richer exploitation of the multimodal framework. [unreadable] [unreadable] [unreadable]

Ammon Horn; Anatomic Models; CRISP; Clinical; Computer Retrieval of Information on Scientific Projects Database; Computer Simulation; Computerized Models; Condition; Cornu Ammonis; Data; Development; Disease; Disorder; Funding; Grant; Hippocampal Formation; Hippocampus; Hippocampus (Brain); Human; Human, General; Image; Individual; Institution; Investigators; MR Imaging; MR Tomography; MRI; MRI Scans; Magnetic Resonance Imaging; Magnetic Resonance Imaging Scan; Man (Taxonomy); Man, Modern; Manuals; Mathematical Model Simulation; Mathematical Models and Simulations; Measurement; Medical Imaging, Magnetic Resonance / Nuclear Magnetic Resonance; Memory; Methods; Methods and Techniques; Methods, Other; Modeling; Models, Anatomic; Models, Anatomical; Models, Computer; NIH; NMR Imaging; NMR Tomography; National Institutes of Health; National Institutes of Health (U.S.); Neurosciences; Nuclear Magnetic Resonance Imaging; Phase; Process; Research; Research Personnel; Research Resources; Researchers; Resources; Role; Scans, MRI; Simulation, Computer based; Source; Techniques; Technology; United States National Institutes of Health; Zeugmatography; base; computational modeling; computational models; computational simulation; computer based models; computerized modeling; computerized simulation; data acquisition; disease/disorder; hippocampal; imaging; in silico; in vivo; in vivo Model; interest; neuroimaging; social role; ultra high resolution; virtual simulation

DESCRIPTION (provided by applicant): Standard structural brain imaging protocols result in images that cannot resolve structures smaller than 1- 2mm in size. Achieving significantly higher resolution would be of fundamental clinical and neuroscientific value, as it would allow the in-vivo detection and analysis of cytoarchitectural features of the cortex, as well as substructures of brain regions such as the hippocampus, thalamus and amygdala. Unfortunately, such resolution is extremely difficult to obtain in-vivo, as the signal-to-noise ratio goes down with the third power of the linear dimension of each voxel. While some recent studies have pushed this limit to under 1A mm, this is at the cost of extremely long scan sessions and specialized imaging hardware, and even this is still a coarse resolution relative to what is required to visualize correlates of the cytoarchitecture with MRI. Here we take a different approach, and propose to image ex-vivo tissue samples, both blocks of tissue and whole hemispheres, in which exceedingly high-resolution is obtainable, on the order of lOOujns. In these images, many MR signatures of cytoarchitectural features are apparent, and hence they can be used for the construction of models including these cytoarchitectonically defined boundaries. For those features that are not distinguishable from the MR, we propose to perform histological analysis of the tissue, and use cross modal registration techniques to transfer the information from the histology back to the models. High dimensional mapping procedures are then proposed to map these models, obtained from ultra high-resolution imaging and histology, back to the more standard resolution in-vivo data to predict the probability of a given cytoarchitectural boundary occurring at each location in the in-vivo data. We focus on cortical areas in the medial temporal lobe as they are of great clinical relevance, as they are thought to be one of the earliest loci of Alzheimer's disease, and are critical to normal memory function. The ability to more accurately localize these cortical regions would be a critical step in the early diagnosis of AD, and in the assessment of the efficacy of potential clinical interventions.

Diffusion tensor imaging can measure the microstructural integrity and orientational structure of nerve tissue. Computational tools to extract microstructural information from DTI will enhance insight into neuroanatomy, neuropathology, and neuropsychiatric disorders. We propose to develop statistical tools to test group-level hypotheses on white matter histoarchitecture based on diffusion imaging. We will extend analytic and statistical tools developed for scalar data, to orientational diffusion data, leading to improved registration, segmentation, and analysis of tissue microstructure.

DESCRIPTION (provided by applicant): This proposal requests funding to support a new, state-of-the-art, high performance storage area network (SAN) for high speed and robust access to large quantities of structural and functional imaging data. Specifically, we propose to obtain a multi-terabyte SAN running a parallel filesystem. This type of high-throughput storage solution is designed to work synergistically with large-scale computational clusters such as our current 340 CPU supercomputer that was funded through a previous shared instrument grant (1S10RR019307-01 A Computeserver For Structural &Functional Image Analysis ). Currently, the Martinos Center's storage is based on traditional single-host systems that run a Network File System (NFS). We have found these systems and NFS unable to scale efficiently to support the hundreds of simultaneous jobs that are frequently run on our computational cluster. In our current architecture, a handful of jobs accessing the same NFS server can cause other jobs to timeout and fail to access the data on that server, resulting in significant difficulties in large-scale neuroimaging studies. We have identified four broad areas of users that will realize dramatic benefit from the proposed instrument: a) morphometry studies (to increase throughput on large N morphometry studies), b)fMRI studies (to increase the throughput of surface analysis and statistical characterization, c) multi-modal integration studies (to facilitate complex forward and inverse modeling used in EEG/MEG/fMRI and diffuse optical tomography), and d) algorithm development users (to allow large parameter spaces to be explored in parallel without requiring multiple copies of data). Each of these classes of users now use operational solutions that depend on ad hoc and lengthy delays between the execution time of sequential jobs, which can dramatically decrease efficiency while offering no guarantee of success. The proposed system is expected to deliver robust and rapid access to storage and obviate the need for tedious and time-consuming rerunning of jobs that fail due to network errors. The broad-based user community, which spans 5 institutions (MGH, Harvard, BWH, MIT, and BU) and many departments within these institutions will benefit greatly from a more advanced storage system and the increased capabilities it engenders. This instrument will enhance the performance capabilities of two Regional Resources, enabling these essential facilities to deliver robust, rapid, and dependable access to data for their users. PUBLIC HEALTH RELEVANCE: The proposed instrument will represent significant enhancement to our current ability to efficiently process large quantities of neuroimaging data. With the large number of imaging studies at the Martinos Center that center on clinical populations for instance, studies of Alzheimer's disease, schizophrenia, Huntington's disease, healthy aging such an instrument thus holds great potential for clear and significant public health benefit. The ability to process large studies in parallel, without engendering failures in file access, will enhance our capacity to detect the subtle, early effects of these types of disorders, a critical capability for the development of proper clinical interventions, which are likely most effective in the early stages of neurodegenerative disorders, before widespread cell death occurs.

DESCRIPTION (provided by applicant): The goal of the current Grand Opportunity is to create a unique, multimodal transcriptional atlas of the pre- and postnatal developing human brain. To achieve this ambitious goal in two years, we propose to bring together expertise in human and non-human primate brain development (Sestan, Yale;Lein, Allen Institute), MRI and DTI imaging (Fischl, Harvard/MGH), large-scale transcriptional profiling (Sestan, Yale;Geschwind, UCLA), and the resources of the Allen Institute for Brain Science and Yale University in bioinformatics, information technology, industrial scale histological data generation and atlas generation, and transcriptome analysis (Lein, Hawrylycz, Allen Institute;Sestan, Gerstein, Yale). Members of this consortium will contribute key data components of the atlas using their existing core capabilities at scale, which will be integrated through a powerful public access web-based portal for viewing, searching and mining of spatiotemporal gene expression patterns in the anatomical context of human brain development. The Allen Institute will create an atlas framework, consisting of a set of de novo multimodal imaging and histological reference atlases spanning human brain development, and a complete informatics structure for mapping, integrating, and presenting transcriptional data in the context of neuroanatomical structure and key developmental events. Contributing laboratories will generate transcriptional data utilizing cutting edge profiling techniques applied at the largest feasible scale, including microarray analysis across the entire fetal brain, RNA-seq deep sequencing data to provide comprehensive genomic coverage in targeted cortical and subcortical structures across 12 developmental stages (Sestan, Yale), and celular resolution data to validate and extend the profiling data and to aid interpretation of dynamic spatiotemporal gene expression patterns. Specifically, the Allen Institute will create or manage generation of 1) 12 new MRI, diffusion tensor imaging (DTI) and histological digital annotated reference atlases spanning human brain development, 2) an anatomically comprehensive microarray analysis of early to mid-fetal brains, with sophisticated visualization and analysis tools and a complementary ISH follow-up data set, 3) a celular resolution ISH atlas of gene expression spanning postnatal brain development, mirroring the ongoing NIH Blueprint rhesus macaque neurodevelopmental atlas, and 4) a new web-based portal for presenting and linking all data modalities, including a massive transcriptome analysis of targeted brain regions spanning human brain development generated by the Sestan laboratory. All of these data will be integrated through the centralized web portal, and linked to a series of similar data resources allowing direct comparative analysis between human, non-human primate and rodent model systems. This portal will create a long-lasting resource for the scientific and medical research communities for relating specific transcriptional programs to processes of human brain development. PUBLIC HEALTH RELEVANCE: The current proposal describes the creation of a new and unique data-rich resource integrating transcriptomic, cellular resolution histology and imaging data in the context of human brain developmental mechanisms through a powerful, easy to use public access web-based portal for viewing, searching and sophisticated mining of spatiotemporal gene expression patterns. The final product, consisting of new reference atlases spanning pre- and postnatal human brain development, systematic and comprehensive transcriptome analysis, large-scale cellular resolution analysis, a variety of tools for visualization and data analysis, and extensive linking to external resources, will create a long-lasting resource for relating specific transcriptional programs to processes of human brain development.

This subproject is one of many research subprojects utilizing the resources provided by a Center grant funded by NIH/NCRR. Primary support for the subproject and the subproject's principal investigator may have been provided by other sources, including other NIH sources. The Total Cost listed for the subproject likely represents the estimated amount of Center infrastructure utilized by the subproject, not direct funding provided by the NCRR grant to the subproject or subproject staff. Models of brain structures generated from magnetic resonance imaging (MRI) data have grown in complexity in recent years, evolving from simple models with few classes such as gray matter, white matter and cerebrospinal fluid (CSF), into more complex ones representing multiple neural structures separately. This evolution has been possible due to developments in MR data acquisition technology that has yielded finer resolution, higher signal-to-noise ratio (SNR) images and an increasing number of contrast mechanisms, all of which have been used by increasingly sophisticated analysis tools to improve and extend classification. Nevertheless, despite these important advances, a critical unmet goal of this type of modeling is the generation of representations of myelo- and cytoarchitectonic boundaries from in vivo imaging data. In this competing renewal we seek to significantly augment and extend the tools developed in the last cycle. Using models and probabilistic information assembled in the previous cycle, we now propose to develop cortical registration and segmentation tools that are explicitly optimal for the alignment and localization of architectonic boundaries across subjects, focusing on cortical area V5/MT. The surface- based registration utility will then be combined with volumetric intensity information to generate a nonlinear volume warp using a biomechanical model of the brain. This combination will yield a single highly accurate coordinate system applicable across the entire brain. Finally, this coordinate system will be used as initialization for extending the scope and level of detail of our existing segmentation models to explicitly segment bone, air, fat and water for used in MR-based PET attenuation correction. The segmentation will be facilitated by the use of specifically designed sequences incorporating ultra-short TE (UTE) contrast that can directly image bone.

DESCRIPTION (provided by applicant): Over the past 10 years we have developed and distributed a morphometry package for automatically characterizing and quantifying neuroanatomical structures in the human brain, including the automated construction of models of the hippocampus, amygdala, ventricular system and neocortex. These tools have been designed for use in a research setting, and have deepened our understanding of a wide variety of neurological disorders and effects such as schizophrenia, Huntington's disease, Alzheimer's disease, Semantic Dementia, phobias, autism, dyslexia, aging, and development. Unfortunately, design decisions made over a decade ago now hamper the adoption of these tools into clinical settings. In this project we propose to use state-of-the art software engineering methods in order to remove these restrictions by designing and engineering these algorithms using current graphics processor unit (GPU) technology, which has been shown to routinely provide on the order of 50-fold speed increases. An additional advantage will be the careful unit testing that can be built into the new tools, ensuring they operate with the high degree of accuracy and reliability required for point-of-care clinical tools. The result will be a suite of open-source algorithms that can run on a standard workstation with a commercially available graphics card, which can rapidly provide diagnostic information for multiple conditions at the point-of-care. PUBLIC HEALTH RELEVANCE: Neurodegenerative disorders result in different patterns of atrophy in the human brain, with each pattern giving rise to the characteristic clinical impairments typically used for diagnosis. Unfortunately, the behavioral changes typically used for diagnosis can be ambiguous, making neuroimaging a potentially valuable diagnostic tool. The goal of this project is to fill this unmet need, by providing a set of tools for quantifying neuroanatomical properties of the human brain from routine clinical MRI scans. Using currently available low- cost graphics cards we will be able to analyze this type of data on standard workstations resulting in information that can be used in diagnosing a wide array of neurological disorders by quantifying the size and shape of structures such as the hippocampus, the amygdala, the ventricular system and the neocortex. We will provide access to statistical information regarding normal variability in these structures, resulting in a tool that can automatically, rapidly and robustly detect abnormal brain anatomy indicative of disease process at the point-of-care.

DESCRIPTION (provided by applicant): The white matter of the human brain is made up of axons that typically travel together in bundles called fascicles. Recent advances in Magnetic Resonance Imaging (MRI) has made it possible to image the self- diffusion of water molecules through Diffusion-Weighted MRI (DWMRI). Because water diffuses significantly more along axons than across them, DWMRI allows one to probe the microarchitecture of the white matter of the living human brain for the first time. In this project, we propose to build models of fascicles by grouping individual fiber models derived from DWMRI data based on the surrounding anatomy. We hypothesize that the anatomy through which a tract passes is predictive of the location of the tract, thus providing information about what fascicle each fiber should be assigned to. The resulting algorithms will provide automated tools for the modeling of the major fascicles in the human brain.

DESCRIPTION (provided by applicant): Human neuroanatomy is enormously variable across subjects - a factor that limits the power of brain studies to detect effects of interest. While degeneration in subcortical structures and cortical gray matter is manifest in many conditions such as aging, Alzheimer's disease, Huntington's disease, multiple sclerosis and schizophrenia, large studies are needed in order to find robust and stable effects that separate groups. Furthermore drug development becomes highly costly as detecting small reductions in atrophy can take years and hundreds or thousands of subjects. These factors raise the importance of longitudinal studies, in which one acquires data at multiple time points and examines the differences in temporal trajectories. Compared to a cross-sectional approach, the longitudinal design can provide more sensitivity and specificity for examining subtle associations by reducing the confounding effect of between-subject variability. Moreover, a serial assessment can be the only way to unambiguously characterize the effect of interest in a randomized experiment, such as a drug trial. Finally, longitudinal studies provide unique insights into the temporal dynamics of the underlying biological process, such as disease progression. Taking full advantage of a longitudinal design requires the optimization of the computational tools that perform image processing and hypothesis testing. In this project, we propose to design, develop and distribute intrinsically longitudinal image processing and hypothesis testing tools and validate them in the study of a set of neurodegenerative diseases.

? DESCRIPTION (provided by applicant): Auto-calibration and shaped insulin delivery to lower average blood glucose Abstract The current clinical standard for calibrating insulin pumps is based on total body weight. Unaccounted for factors in this procedure result in estimates that are frequently incorrect by 25-30%. The result is that in most cases months of trial-and-error are required to arrive at a reasonable pump calibration. This is particularly problematic in children, as their pumps require re-calibration whenever they grow significantly or rapidly (e.g. during growth spurts). The proposed algorithm would compute individual, accurate values for calibrating pumps, automatically setting all the requisite parameters (insulin sensitivity, carb ratio and basal rates) from information regarding carb intake and a few measures of blood glucose level. A second advance in the proposed project would be to use complex, time-varying insulin doses to minimize postprandial high and low blood glucose excursions. This project would shift some of the basal insulin that would otherwise have been given over the hours subsequent to the ingestion of carbohydrates to either before or during the meal, resulting in a dramatic lowering of postprandial blood sugars without the risk of hypoglycemia.

? DESCRIPTION (provided by applicant): The human brain is made up of an array of functionally and structurally defined regions. Localizing these regions is critical for early diagnosis of an array of diseases such as Alzheimer's disease (AD), as well as for neuroscientific research aimed at understanding the brain's functional and structural properties and clinical intervention in drug-resistant depression. Unfortunately, direct visualization of the laminar properties that are one of the defining characteristics of cortical areas is beyond current imaging technology, except in specialized cases that represent a small fraction of the entire cortex. In this project we propose to develop tools for segmenting cortical areas and laminar boundaries using ultra-high resolution ex vivo MRI and optical coherence tomography (OCT), then to make in vivo inferences via probabilistic modeling. This will allow us to develop and distribute tools that should permit automated areal and laminar localization, facilitating an array of clinical and neuroscientific research. The proposed analysis stream holds the promise of automatically deriving new laminar and area-specific morphometric measures, a critical advance in our goal of detecting neurodegenerative disorders early in their course, when therapeutic intervention is still possible, as well as for guiding disease treatment and staging in individual patients suffering from diseases such as depression or multiple sclerosis.

? DESCRIPTION (provided by applicant): Abstract While histology remains the gold standard for assessing human neuroanatomy, the procedures for sectioning and hand mounting tissue for microscopic imaging are not substantially different than they were 100 years ago. These steps introduce irremediable distortions into the tissue sections making it difficult or impossible to align sections with suffiient accuracy to create 3D histological volumes at the micron scale. In this project, we seek to develop acquisition and analysis tools that use optical coherence tomography (OCT) to generate images that contain information comparable to standard histology. Critically, OCT images the tissue prior to cutting, thus avoiding the concomitant distortions, and allowing large regions of human tissue to be imaged with micron resolution. We anticipate that these large-scale, veridical representations will facilitate the development of automated techniques for tissue quantification and disease detection, dramatically increasing the efficiency, specificity an sensitivity of histology and neuropathology.

Abstract: FreeSurfer Development, Maintenance, and Hardening Imaging of the human brain has seen explosive growth in the last decade mainly through the various modalities of MRI. The massive amount of data requires automatic and robust tools for analysis. FreeSurfer (FS, surfer.nmr.mgh.harvard.edu) is one of the preeminent tools used for neuroimage analysis. FS has more than 20,000 downloads, and the core FS manuscripts have been cited more than 3000 times. FS is part of the analysis core for many NIH-funded large-scale data acquisition projects such as the Human Connectome Project (HCP), Alzheimer's Disease Neuroimaging Initiative (ADNI), and Frammingham Heart Study (FHS). One third of all ADNI-based publications cite FS. Simply put, much of the innovative research done in neuroimaging would not be possible without FS. Started in 1998, FS is best known for providing detailed and automated anatomical analysis of T1- weighted MRI images, especially for the cortical surface. However, FS anatomical analysis provides an ideal substrate for all modes of brain imaging including functional MRI, diffusion MRI, PET, optical/NIRS, EEG/MEG. FS provides tools to perform these analyses as well as software to integrate with other analysis tools (eg, SPM, FSL, AFNI). FS has been used for presurgical planning and even in the operating room. A software package with a scientific breadth and user based the size of FS?s requires a significant amount of effort just to maintain it. For example, the FS email list receives approximately 3000 posts a year. FS must be continuously and rigorously tested because it is such an integral part of the neuroimaging infrastructure. Users are constantly requesting new functionality and better performance. This proposal will be used to develop, maintain, and harden FS. Specifically, we will make FS more robust by incorporating multiple modalities instead of just T1. We will replace the whole-brain segmentation with an unsupervised method that simultaneously optimizes bias field correction in a multimodal setting. We will implements multivariate analysis tools to assist in the interpretation of data. We will also harden and optimize the FS code base. Finally, we will include tools to assist the user to easily find where the FS analysis fails.

Abstract: FreeSurfer Development, Maintenance, and Hardening Imaging of the human brain has seen explosive growth in the last decade mainly through the various modalities of MRI. The massive amount of data requires automatic and robust tools for analysis. FreeSurfer (FS, surfer.nmr.mgh.harvard.edu) is one of the preeminent tools used for neuroimage analysis. FS has more than 20,000 downloads, and the core FS manuscripts have been cited more than 3000 times. FS is part of the analysis core for many NIH-funded large-scale data acquisition projects such as the Human Connectome Project (HCP), Alzheimer's Disease Neuroimaging Initiative (ADNI), and Frammingham Heart Study (FHS). One third of all ADNI-based publications cite FS. Simply put, much of the innovative research done in neuroimaging would not be possible without FS. Started in 1998, FS is best known for providing detailed and automated anatomical analysis of T1- weighted MRI images, especially for the cortical surface. However, FS anatomical analysis provides an ideal substrate for all modes of brain imaging including functional MRI, diffusion MRI, PET, optical/NIRS, EEG/MEG. FS provides tools to perform these analyses as well as software to integrate with other analysis tools (eg, SPM, FSL, AFNI). FS has been used for presurgical planning and even in the operating room. A software package with a scientific breadth and user based the size of FS?s requires a significant amount of effort just to maintain it. For example, the FS email list receives approximately 3000 posts a year. FS must be continuously and rigorously tested because it is such an integral part of the neuroimaging infrastructure. Users are constantly requesting new functionality and better performance. This proposal will be used to develop, maintain, and harden FS. Specifically, we will make FS more robust by incorporating multiple modalities instead of just T1. We will replace the whole-brain segmentation with an unsupervised method that simultaneously optimizes bias field correction in a multimodal setting. We will implements multivariate analysis tools to assist in the interpretation of data. We will also harden and optimize the FS code base. Finally, we will include tools to assist the user to easily find where the FS analysis fails.

3-Dimensional; Algorithms; Alzheimer's Disease; Architecture; Area; Atlases; Autistic Disorder; Back; base; Brain; brain cell; Brain region; Cell Count; Cell physiology; cell type; Cells; Censuses; Characteristics; Classification; cohort; Data; density; design; developmental disease; Diffusion Magnetic Resoce Imaging; Disease; Dyslexia; Face; Fiber; frontal lobe; histological image; histological stains; Histological Techniques; Human; human tissue; Image; Image Analysis; imaging study; Imaging technology; Impairment; in vivo; in vivo imaging; Individual; Infrastructure; Language; Magnetic Resoce Imaging; Maps; Microscopic; microscopic imaging; Modeling; Molecular; molecular marker; Myelin; Neuroglia; neuroimaging; Neurons; Neuropil; Optical Coherence Tomography; Procedures; Property; Protocols documentation; Resolution; Sampling; scale up; Schizophrenia; Slice; Stains; Structure; System; Techniques; Technology; Thick; Three-dimensional analysis; Tissues; tractography;

Abstract Magnetic Resonance Imaging (MRI) is a flexible and powerful technology for quantifying the effects of many conditions, including numerous neurological disorders, on human brain anatomy, connectivity, vasculature, chemical composition, physiology and function. In the past 15 years, several open source tools have been developed that accurately and automatically segment an array of brain structures. In this project, we seek the resources to extend this set to enable the quantification of neuroanatomical changes that are critical to diagnosing, staging and assessing the efficacy of potential therapeutic interventions in diseases such as Alzheimer?s and Parkinson?s. This includes the acquisition of datasets to enable manual labeling of structures of interest, the generation of documentation, tutorials, unit tests, regression tests and system tests to harden the tools and make them usable by clinicians and neuroscientists, and finally the distribution and support of the data, manual labelings and tools to the more than 32,000 researchers that use FreeSurfer through our existing open source mechanism.

Abstract FreeSurfer is a tool for the analysis of Magnetic Resonance Imaging (MRI) that has proven to be a flexible and powerful technology for quantifying the effects of many conditions, including numerous neurological disorders, on human brain anatomy, connectivity, vasculature, chemical composition, physiology and function. In the past 20 years, these open source tools have been developed to accurately and automatically segment an array of brain structures and have become the core analysis infrastructure for the Alzheimer?s Disease NeuroImaging Initiative (ADNI). In this project, we seek the resources to radically increase the speed, accuracy and flexibility of these tools, taking advantage of exciting new results in Deep Learning. This will enable us to more accurately quantify neuroanatomical changes that are critical to diagnosing, staging and assessing the efficacy of potential therapeutic interventions in diseases such as Alzheimer?s. This includes the generation of documentation, tutorials, unit tests, regression tests and system tests to harden the tools and make them usable by clinicians and neuroscientists, and finally the distribution and support of the data, manual labelings and tools to the more than 40,000 researchers that use FreeSurfer through our existing open source mechanism. In addition, we will analyze the entire Alzheimer?s Disease NeuroImaging Initiative dataset and return it for public release, including a set of manually labeled data that can be used to optimize Deep Learning tools for Alzheimer?s Disease over the next decade.

Project Summary (maximum 30 lines) The BRAIN initiative seeks to develop and apply technologies in order to understand of how brain cells interact in both time and space to give rise to brain function. A key deliverable in BRAIN is a systematic census of neuronal and glial cell types, which is a prerequisite to understand how these cells interact and change in healthy aging and in disease. Moreover, such a census will provide a common reference cell taxonomy, which is crucial to harmonize studies at different sites and achieve the goals of BRAIN. A necessary companion of the census is a reference coordinate system, which enables us to understand the spatio-anatomical context in which cells interact, as well as their connectivity. Building such a coordinate system requires advanced spatial alignment (registration) tools, since virtually every lab technique used in microscopic brain cell phenotyping ? particularly in human brain ? requires blocking and/or sectioning of samples, hence distorting the structure of tissue. Due to the difficulty of providing support for datasets and acquisition setups different to the original, most publicly available techniques to recover the lost tissue structure (?3D reconstruction?) rely on very simple techniques, such as vanilla pairwise registration of neighboring sections. Moreover, conventional reconstruction methods are notoriously slow, and no available method is designed to 3D reconstruct whole human brains. In this interdisciplinary project, which lies at the nexus of computer science, MRI physics, histology, optical imaging, anatomy and statistics, we propose to extend, robustify, test, distribute and support our recently developed, state-or-the-art techniques that will enable the constructions of a coordinate system capable of representing multi-scale maps of human brain anatomy and function. This includes algorithms and software for: image analysis of ex vivo MRI; construction of laminar models of the human cerebral cortex; 3D reconstruction of microscopic images and alignment to the laminar models; surface based analysis of microscopy data on the laminar structure; and alignment of ex vivo and in vivo images to accurately transfer information from microscopy to MRI studies of the living brain, in health and in disease. The tools we propose to build and disseminate will combine modern deep learning techniques with principled Bayesian inference, and have the potential to deliver accurate registration at the macroscopic, mesoscopic, and microscopic level, with high throughput delivered using cutting-edge machine learning algorithms. Effective dissemination of these tools, along with companion test data, will be achieved through our widespread package FreeSurfer. The distributed tools will not only enable the construction of a cell census with rich spatial information at human brain scale (including a novel laminar model), but will also have a tremendous impact in other areas of neuroimaging, including overarching goals of BRAIN such as: linking cellular-level activity to functional MRI, atlas building, or connecting axonal anatomy to diffusion MRI.

Abstract - The vast majority of information that neuroscience has obtained about the microscopic structure of the human brain ? the substrate for cognitive competencies and the specific locations of neuropathological processes ? has been obtained by the analysis of ex vivo tissue. Historically this involves the decades (if not centuries) old procedure of cutting, staining, mounting and imaging under a microscope. The last two decades have seen stunning advances in imaging and analysis of the human brain. This include advances in microscopic (e.g. CLARITY1, SWITCH), mesoscopic (e.g., polarized light imaging, PLI), optical coherence tomography (OCT), RNA-seq and macroscopic imaging (e.g., MRI). While these techniques have generated huge amounts of new information regarding the structural, molecular, connectomic, genetic and transcriptional nature of the brain, they have thus far had little impact on in vivo analysis. In the same way, while we have made great progress in our ability to localize important brain regions in living subjects, these capabilities have had little impact in microscopy and neuropathology. In this project we seek to use our mesoscopic imaging and analysis tools to remove these barriers and facilitate the flow of information from microscopy to in vivo human studies, as well as in the reverse direction. Examples of the impact of these new abilities would be: using resting-state fMRI networks (rsFMRI) to guide the extraction of neuropathological blocks during autopsy to test network-based theories of various neurodegenerative disease or using predicted vascular distributions and densities to improve the laminar specificity of fMRI.

Abstract The ~86 billion neurons that form the human brain are organized at multiple scales, ranging from the fine details of an individual neuron?s dendritic arborization, to local circuits that are embedded within large-scale systems spanning the brain. In this project, we will image across this vast range of scales to create a multiscale atlas akin to Google Earth for the human brain that can visualize hemisphere-wide networks and then zoom in to see individual, labeled cells at micron resolution in the frontal temporal lobe. This dramatic advance will be made possible through the use of an array of imaging technologies, including light-sheet microscopy (LSM), tissue clearing, immunohistochemistry (IMH), magnetic resonance imaging (MRI) and newly developed techniques in Optical Coherence Tomography (OCT). OCT in particular is a potentially transformative technology as it provides micron resolution over large volumes of tissue, images all of the tissue (as opposed to fluorescence), does not require mounting and staining and hence can be automated, and is essentially distortion free as it images the tissue prior to cutting. LSM-based IMH will provide molecular, morphological and spatial properties of cells that will enable us to develop cellular classification systems, while OCT images of the same tissue will enable us to remove the distortions induced by cutting and clearing, and transfer information to whole-hemisphere MRI for atlasing and in vivo inference. This transfer of information depends critically on the ability to register images across a huge range of resolutions and contrast types. For this we propose to use the endogenous fiducial landmarks provided by the cerebral vasculature. To take full advantage of the vasculature using deep learning requires a training set of labeled vessels in each of our imaging modalities across an array of examples. The goal of this supplement is to provide these labelings including the assessment of intra- and inter-rater reliability.

Abstract: Imaging of the human brain has seen explosive growth in the last two decades mainly through the various modalities of MRI. The massive amount of data requires automatic and robust tools for analysis. FreeSurfer (FS, surfer.nmr.mgh.harvard.edu) is one of the preeminent tools used for neuroimage analysis. FS has more than 44,000 downloads, and the core FS manuscripts have been cited more than 22,000 times. FS is part of the analysis core for many NIH-funded large-scale data acquisition projects such as the Human Connectome Project (HCP), Alzheimer's Disease Neuroimaging Initiative (ADNI), Framingham Heart Study (FHS), The Adolescent Brain Cognitive Development (ABCD), as well as the UK BioBank. One third of the 600+ ADNI-based publications cite FS. Simply put, much of the innovative research done in neuroimaging would not be possible without FS. Started in 1998, FS is best known for providing detailed and automated anatomical analysis of T1-weighted MRI images, especially for the cortical surface. However, FS anatomical analysis provides an ideal substrate for all modes of brain imaging including functional MRI, diffusion MRI, PET, optical/NIRS, as well as EEG/MEG. FS provides tools to perform these analyses as well as software to integrate with other analysis tools (e.g., SPM, FSL, AFNI). FS has been used for presurgical planning and even in the operating room. The original grant mostly centered around Sequence Adaptive Multimodal Segmentation (SAMSEG). SAMSEG uses parametric Bayesian generative modeling to segment brain images. The SAMSEG framework fits atlas priors and multivariate Gaussian intensity models to brain images (including MRI artifacts such as bias fields). SAMSEG can take any modality or combination of modalities as input. Since it adapts its intensity model, it is robust to differences in scanner. Since it is a generative model, it is easy to extend to encompass more segmentation details. For example, the SAMSEG framework has been used to segment hippocampal subfield, amygdalar nuclei, thalamic nuclei, and extracerebral structures. The main vision for the renewal is to extend the SAMSEG framework to accommodate longitudinal models, incorporate more anatomical details, and to use SAMSEG output as a basis for cortical surface placement that is, like SAMSEG, modality independent and capable of using any combination of modalities. In addition, we propose a series of new tools that will assist in the individual and group analysis of large studies by creating study-specific models. In addition to this new technical development, we are requesting support for software engineering, maintenance, and user support ? mundane and not innovative, but high-impact this type of support is critical to the thousands of researchers who rely on FreeSurfer.