Digital atlases provide semantic and spatial information that can be used to link together rich collections of data from disparate sources [1
]. The most common types are anatomic atlases that spatially delineate and semantically label the structures of a volumetrically imaged subject. These anatomic atlases can be used as templates for identifying regions of interest in non-delineated data sets, such as localizing gene expression in the brain as visualized by in situ hybridization [2
]. Atlases can also be used to unify implicitly associated data. For example, one data source may use an anatomical name for a location while another uses image coordinates. By aligning both data to a common atlas, correlations can be made. As the quality and amount of biological data continues to advance and grow, having the ability to search, reference, and compare this data with a researcher's own data is essential.
Central to integrating and correlating the continually growing volumes of distributed and online data is the atlas analysis workflow. As shown in Figure , the atlas analysis workflow consists of three major steps: (1) searching and retrieving source data, (2) aligning the source data to the atlas, and (3) comparing and correlating the data in context of the atlas space. This process can be repeated multiple times as interesting or unexpected correlations spark new ideas and findings.
Figure 1 Digital Atlas Analysis Workflow. In the first stage, a researcher searches and retrieves data to help answer the proposed question. The data can be of different types and from multiple sources. If necessary, the data is aligned to a common space, either (more ...)
Several challenges exist with the digital atlas workflow that make conducting the analysis a cumbersome and time-consuming process. First, each data source typically has a different search interface and result format which makes reconciling data difficult. Users must use different interfaces, such as visiting separate web sites, and utilize different software tools to view and analyze the data. Second, spatially aligning or registering image data to an atlas or to each other is a complex problem with many solutions. Factors that influence what type of registration algorithms to use include image modality of the data, dimensionality of the source and template data (2D-2D, 3D-3D, or 2D-3D), speed or time requirements, and accuracy, such as whole brain or structure accuracy [4
]. Making assumptions about any of these factors limits the generality and usefulness of the overall application. Third, although visually comparing and contrasting the data is a natural and intuitive method of analysis, few computer applications can integrate image data, numerical data, annotations, ontological relationships, and atlas data in a meaningful manner.
To address these issues, this work presents the MouseBIRN Atlasing Toolkit (MBAT) - a cross-platform, free open-source software tool designed to accelerate the timeline for integrating and correlating biological data. MBAT empowers researchers to discover correlations among disconnected and disparate data by providing a unified environment for bringing together distributed reference resources, a user's image data, and biological atlases into the same spatial or semantic context. In a single application, MBAT provides the ability to use a single query to search and retrieve data from multiple data sources, align image data using the user's preferred registration method, composite data from multiple sources in a common space, and link relevant information to the current view of the data or atlas. Through its extensible tiered plug-in architecture, MBAT allows researchers to customize all application components to quickly achieve personalized workflows.
As shown in Table , related cross-platform, free open-source software include an atlas navigator, JAtlasViewer [5
], and two medical imaging analysis tools with plug-in architectures, Medical Imaging Processing Analysis and Visualization (MIPAV) [6
] and Slicer [7
Comparison of related software.
JAtlasViewer is a lightweight Java application that provides arbitrary section views of 3D image data, 3D surfaces of pre-segmented and labeled anatomy, and anatomical browsing. Only a single data set can be loaded at a time so it lacks the ability to perform comparisons among data sets. It also lacks any search and registration features.
MIPAV and Slicer are more mature, larger efforts to build a suite of tools for medical image analysis. Both provide powerful visualization, registration, and analysis features. However, only Slicer provides some limited labeling and search functionality. In the Volume module, the label volumes can be overlaid on image volumes but only the label id can be interrogated; the corresponding label name and color are not supported. In the QueryAtlas module, label names and colors are displayed on 3D surfaces or sectional views, with links to several ontologies, but the query engine only supports literature sources. While MIPAV supports plug-ins, the types of plug-ins are limited to image processing algorithms, file format readers, and rendering methods. Slicer's plug-ins, called modules, offer greater flexibility as the modules can be implemented as shared libraries or leverage existing legacy executables. To simplify the development, Slicer also provides methods to auto-generate GUIs for the plug-ins. However, Slicer's architecture is based on manipulating objects in a global 3D scene so adding non-graphical objects, such as search objects, to Slicer is cumbersome.
What sets our work apart from the others are three key features. First, our framework has the ability to integrate multiple data types from multiple, disparate sources. Gathering and searching for data must be performed external to the other applications. Second, our framework adopts the plug-in architecture from top to bottom and is designed for interoperability. Parts of other applications can be integrated into our application and vice versa, parts of our application can be integrated into other applications. Third, novel digital atlasing features are developed, such as supporting multiple label sets, dynamic selection and grouping of labels, and synchronized display of ontological data.