Combining image content and annotated text for medical image categorization and retrieval

The richness of health-information available on-line requires the development of efficient information retrieval methods. The CISMeF heath-catalogue provides indexing and searching capabilities for healthresources. Medical images are representing a significant part of on-line medical knowledge and a valuable component of diagnosis and teaching. In this context, a combined text and image extraction approach is desired for the indexing and retrieval of on-line medical information. This paper presents the capabilities of the MedIC system of automatically extract image information from the image visual content or the image annotated text, as well as the current architecture for combining the decisions issued at each of these levels, using Bayesian Networks. The results we are presenting are showing that combining these decisions can create a better result. Extracting further image-related information from the documents where the images are placed, will further test the this information fusion approach, and ultimately enrich the MedIC image categorization module, of the CISMeF catalogue.