Image matching as a diagnostic support tool

Abstract

We have developed and evaluated a novel image-matching method for medical images. This method allows the radiologist to search through - in a matter of seconds - large medical databases containing thousands of patients. To illustrate the usefulness of this method in a clinical setting, we have employed this method as a diagnostic support tool for pediatric brain diseases. To this aim, we have assembled a database containing Magnetic Resonance (MR) brain images of 2500 patients between ages 0 and 18 with known brain lesions. As the images are added to the database, they are registered to a global coordinate system. In addition, regions of interests (ROI) are labeled, and sophisticated image processing techniques are used to extract image parameters from the ROIs and from the entire MR image. To perform a clinically realistic search through this database, we have established a training testbed at Childrens Hospital Los Angeles for acquiring MR images from our PACS server of patients with unknown lesions. We have matched these images with the images in the pediatric brain MR database containing known lesions using our image-matching method. An expert pediatric neuroradiologist evaluated the search results. We found that in most cases, our image-matching method is able to retrieve images with relevant diagnostic content, making it highly attractive as a diagnostic support tool.