Global Exploratory Analysis of Massive Neuroimaging Collections using Microsoft Live Labs Pivot and Silverlight

Abstract

Event Abstract Back to Event Global Exploratory Analysis of Massive Neuroimaging Collections using Microsoft Live Labs Pivot and Silverlight Robert W. Williams1*, Lei Yan1, Xiaodong Zhou1, Lu Lu1, Arthur Centeno1, Leonard Kuan2, Michael Hawrylycz2 and Glenn D. Rosen3 1 University of Tennessee Health Science Center, Department of Anatomy and Neurobiology, Center for Integrative and Translational Genomics, United States 2 The Allen Institute for Brain Science, United States 3 Beth Israel Deaconess Medical Center/Harvard Medical School, Department of Neurology, United States Introduction: The Mouse Brain Library (www.mbl.org) consists of high resolution images generated for 200 genetically well defined strains of mice, 2200 cases, 8800 Nissl-stained slides, and ~120,000 coronal and horizontal sections of the brain. The MBL includes representatives for most sets of recombinant inbred strains, (BXD, LXS, AXB, BXH, CXB). This collection provides a solid foundation for studies of the genetic control of brain structure, function, and behavior (Lu et al. 2001; Yang et al., 2008; Rosen et al., 2009). A key challenge is how to deliver massive image collections such as the MBL, Allen Brain Atlas projects, and BrainMaps using modern web services. In this work we tested Microsoft Live Labs Pivot (www.getpivot.com ) as an image distribution portal. Pivot provides a unique method for delivering and exploring very large collections of high-resolution images. Methods and Results: Large slides of 30-µm celloidin sections were imaged using a 20x objective (NA 0.75) and an Aperio Scanscope CS scanner at an in-plane resolution of 1-µm/pixel and at 150-µm steps along the cutting axis. Images were segmented to extract single sections. SVS format image were converted into Advanced Forensic format (AFF), JPEG2000, and Silverlight deep zoom (DZI) pyramids. A DVI is 2x larger than its JPEG parent. DZIs were assembled into a set of Pivot image collections. Metadata for sorting and display were extracted from MBL and GeneNetwork.org data tables. Metadata types include thousands of genotypes and phenotypes for all strains, as well as data for individual cases (sex, age, body weight, brain weight, litter size). Spatial coordinate tags for each section using INCF Waxholm space and BIRNLex neuroanatomical terms are being added. A Pivot server was installed on a Linux CentOS 5 8-core system (Dell R610) running Tomcat and MySQL. The system was tested on Vista and Windows 7 using a Pivot browser (www.getpivot.com) and runs well on Mac OS X 10.6 using the VMWARE Fusion 3 virtual machine. Link to http://mbl.pivotcollections.org to view the MBL. The MBL collection is large and unwieldy and our current interface (www.mbl.org) does not provide sufficient sorting flexibility and speed to effectively explore or analyze the collection. A good exploratory interface would provide both forest and tree views and a way to effectively scan the entire collection for variation in traits such as ventricular volume, callosal architecture, cortical lamination, and differences in the cytology in specific nuclei in a matter of minutes, not hours or days. Pivot is optimal for these types of rapid review and exploratory tasks. The collection can be filtered, sorted, and viewed at a wide range of magnifications—from thumbnails of whole slide to full zooms of subnuclei—almost instantly. The collection can be split, filtered, and sorted using a range of continuous and discrete metadata variables (sex, age, strain, genotype, behavior) Limitations with the current Pivot implementation can be divided into two categories: those associated with the interface itself (no nested displays, a limit of 15 displayed categories, no graphic overlay for marking or annotation), and those associated with secondary analytic and statistical functions that would typically be used to test hypotheses (no dynamic output of group statistics nor tests for differences among groups using ANOVA or t tests) Discussion: Pivot is a superb web service architecture that provides very fluid access to massive neuroimaging databases. It is extremely well suited for both the dissemination of massive 2D collections and direct exploratory analysis as part of a web service. Pivot has also been extremely helpful as part of quality control workflow and has enabled us to search for neuroanatomical differences and patterns of variation among strains of mice in ways that far surpass any other web interface.