The ability to search for visually similar images in a PACS database would be helpful in both research and diagnostic applications, and this benefit can be provided by combining content-based image retrieval techniques with text retrieval methods, according to a presentation by researchers from the Shanghai Institute of Technical Physics in China.
Due to technical challenges in working with large databases, the use of content-based image retrieval techniques has been limited in medical applications. But applying text retrieval technology -- specifically, a semantic space-searching method -- with these techniques can solve those problems, utilizing information from both images and radiology reports for indexing and retrieval of images in a RIS/PACS environment, according to Jianguo Zhang, Ph.D., professor and director of the institute's Laboratory for Medical Imaging Informatics.
"We combine text retrieval technology (which is widely used by Google and Yahoo) and [content-based image retrieval] techniques to improve the performance and precision of the image retrieval process," according to Zhang. "This approach may be used to process specific disease images combined with related reports to make [content-based image retrieval] functionality practically used in PACS and RIS."
The research team said it has tested its method in its clinical practice on 150 cases of lung CT images with solitary pulmonary nodules (SPN) and their associated radiology reports. Search results with high levels of image similarity and the same semantic descriptions are likely to have the same pathologies as the example images, offering value in the diagnostic process, according to the authors.
