IMAGE FILE FORMATS

Source image data formats vary from institution to institution and are often heterogeneous even within a single research group. The radiology community has made great progress in developing and promulgating the DICOM standard, but each equipment vendor has a slightly different implementation of the standard and compatibility problems are still common. Historically, research-oriented image analysis programs have included file import routines to convert a subset of the possible input data formats into the applications own particular data format. Often, information is discarded during this conversion process so that, for example, the data relating to the scan parameters during the image acquisition are lost. In the context of the BIRN database, which is intended to support a wide range of possible research applications, such truncation of the data is not desirable. Consequently, we are developing a set of data conversion routines to bring the source images into standard DICOM with attention to making the resulting files as widely compatible as possible and retaining the full data associated with the original files.

DATA FILE FORMATS

Data file formats for the representation of research results face a similar set of issues to those described above. However, there is much less standardization in this area and as a result each analysis application is likely to have developed an ad hoc method of representing the associated data. Within this category, typical data types include

• images (either modified versions of source images or image masks representing anatomical labels),
• geometric information such as 3D models of anatomy, or
• volumes of scalar or vector valued fields.

It is sometimes the practice to store this information in industry-standard file formats. For example, storing all image data in DICOM formatted files or adopting CAD file formats for geometric model data. This problem has begun to be addressed through the development of meta-data description systems such as the Medical Reality Markup Language (MRML). This type of approach addresses the problem of heterogeneous data representation through the addition of a middleware layer of software that helps support interpretation of different data formats. A goal of the development associated with the BIRN initiative is a detailed exploration of existing practices within the partner sites in order to identify software architectures that have proven effective at addressing the data representation problem.

ANALYSIS SOFTWARE

The final major area of development, and perhaps the most challenging, is the issue of run-time interoperability of the various analysis software applications. Specifically, we are referring to the ability to mix and match software modules from the various BIRN partner sites as needed to support a given research scenario. Each of BIRN sites has extensive experience within their institution on these issues. The BIRN project is focused on exploring each of these options and developing a set of software tools to facilitate the adoption of the best-demonstrated practices. BIRN supports both a scripting interface, in which stand-alone application modules communicate through intermediate files under the control of higher level control program, and a shared library interface in which routines are loaded into a single application context and operate directly on the same data structures in memory.

In This Section

• Function BIRN
  
  
• Morphometry BIRN
  
  
• Mouse BIRN
  
  
• BIRN Coordinating Ceter