Advances in imaging have had a profound effect on our ability to generate high-resolution measurements of the brain?s structure. One of the major hurdles in processing modern neuroimaging datasets designed to produce large-scale maps of the connections and the organization of the brain lies in the sheer size of these data. For instance, electron microscopic (EM) images of a cubic millimeter of cortex occupies roughly 3 PBon disk, and lower resolution emerging X-ray microtomography (XRM) data can exceed 10 TB for a single mouse brain. When dealing with datasets of this size, the application of even simple algorithms becomes difficult. The size of datasets also exacerbates the considerable challenges for dissemination, reproducibility, and collaboration across laboratories. Addressing these challenges requires a new approach that leverages state-of-the-art computer science technology while remaining conscientious of the underlying bioinformatics. We propose Scalable Analytics for Brain Exploration Research (SABER), a user-friendly and portable framework that automates the retrieval, extraction, and analysis of large-scale imagery data to facilitate neuroscientific analyses.
SABER aims to improve the reliability and reproducibility of neuroimagery research by providing a common substrate upon which algorithms may be developed. Leveraging SABER?s containers ? a standardized packaging for software ? this substrate can then be trivially transferred to other machines by the same researcher or by other teams aiming to reproduce or adapt the prior work, making sharing workflows and extracting knowledge commonplace. Using SABER will ensure that the analysis runs identically, regardless of by whom or where the workflow is executed. Because developing and deploying these analysis solutions for large image volumes are acute barriers to developing consistently reproducible workflows, SABER will further the neuroscientific analysis community by simplifying the workflow-development and workflow-execution steps. To demonstrate this, we plan to distribute two community-vetted, optimized workflows to convert large-scale EM and XRM volumetric imagery into maps of neuronal connectivity. Many neurological diseases are characterized by their impact on the density of cells and vessels, neuron death, connectivity, or other factors that are visible with imaging technologies. SABER will provide a framework for producing reproducible estimates of cell counts, vasculature density, and connectomes, thus enabling increased understanding of the impact of disease on the neuroanatomy of many brains. This work will enable the development of tools that can both be applied to massive data and shared amongst many scientists, which will in turn accelerate progress and neuroscientific discovery.
Our Johns Hopkins University Applied Physics Laboratory team leverages prior neuroscience analysis experience to present SABER: Scalable Analytics for Brain Exploration Research ? a portable, easy-to-install framework that enables large-scale neuroanatomical data processing by providing a scaffold upon which highly-reproducible bioinformatics protocols may be built. To support emerging efforts to understand the biological basis of disease, we demonstrate turn-key pipelines to translate multi-terabyte electron microscopy and X-ray microtomography data volumes into maps of neuronal connectivity.
| Yang, Xiaogang; De Andrade, Vincent; Scullin, William et al. (2018) Low-dose x-ray tomography through a deep convolutional neural network. Sci Rep 8:2575 |
| Matelsky, Jordan; Kiar, Gregory; Johnson, Erik et al. (2018) Container-Based Clinical Solutions for Portable and Reproducible Image Analysis. J Digit Imaging : |
