Michael Halle PhD, Abstract Co-Author: Nothing to Disclose

Samira Farough MD, Presenter: Nothing to Disclose

Marianna Jakab MSc, Abstract Co-Author: Nothing to Disclose

Ron Kikinis MD, Abstract Co-Author: Nothing to Disclose

We believe that the widespread availability of ontology-augmented atlases and labeled image data will encourage new innovation in the field of medical image analysis, including the development of new anatomy-specific medical image processing algorithms, user interfaces, and rich teaching tools.

RadLex provides an ontological representation of formal medical and radiological knowledge. To promote the use of RadLex in imaging, we have made publicly available several multi-modality, multi-format image atlases of normal morphology with links to RadLex identifiers and metadata. Three atlases are available: an MRI-based brain atlas of a normal subject consisting of T1- and T2-weighted acquisitions at 0.75mm isotropic resolution, a CT-based abdominal atlas, and a MRI-derived knee atlas. The atlases include original scan data, per-voxel labels of anatomic structures, geometric models of those structures, links between structures and RadLex, and scenes for viewing the data in the medical visualization package 3D Slicer. Information about the atlases can be found at http://nac.spl.harvard.edu/pages/Atlases

Development of these atlases is motivated by the more than 13,000 downloads of previous atlases from our site since 2008. Raw scan data for each atlas is first segmented using a mature image processing pipeline. To insure greatest accuracy, this initial segmentation is then edited by a physician using manual outlining and voxel-by-voxel editing. The physician chooses structure names and RadLex IDs by referencing ontology terms and synonyms. RadLex is then used to generate physician-reviewed structural hierarchies and views. In the process of atlas development we have reported more than twenty corrections to RadLex. These minor changes were included in subsequent RadLex releases.

RadLex augmentation enables useful operations on atlas image data. For example, RadLex relations such as part, system, or organ type can be used to select image regions for processing or display. Organ groups or systems can be extracted. Standardized naming aids in associating atlas information to patient-specific data. Benefits increase as new knowledge is added to RadLex.

Halle, M, Farough, S, Jakab, M, Kikinis, R, Publicly Available RadLex-linked Anatomy Atlases for Image Analysis, Informatics, and Education.  Radiological Society of North America 2011 Scientific Assembly and Annual Meeting, November 26 - December 2, 2011 ,Chicago IL. http://archive.rsna.org/2011/11009748.html