Content-based image retrieval with semantic navigation for medical images with multifocal diseases in integrated RIS/PACS system
Yanjie Zhu, Jianguo Zhang · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2011
In this paper, we proposed a novel architecture integrated with RIS/PACS system that combined image annotation, CBIR techniques and high-dimensional index to retrieve similar medical images with one or more relevant focus in large scale medical image database. In our designed system, regions of interest (ROIs) were labeled by symptom descriptions found in relevant radiology reports as semantic navigation. The annotations were saved as xml file with image makeup language (IML). Then low level features such as texture and statistic features were extracted from the ROIs of lesions and inserted into a database. Recursive feature elimination algorithm was applied to find a high performance feature subset for each symptom. These subsets were used to build high dimensional index with semantic labels guiding the searching path as the navigation. As there might be more than one focus in one image, weight values specified by the user were introduced to calculate the final similarities. The searching results of medical images with multi-focal diseases are likely to have the same pathologies and visual effects with example image and are valuable for imaging diagnosis. The system was implemented for lung CT images, but it could be easily extended to other organs.