Efficient and Effective Matching of Compound Patient Records

Michael Springmann, Sören Balko, Hans-Jörg Schek

Workshop Working Notes

2005/3

8th DELOS thematic workshop - Future Digital Library Management Systems:System Architecture and Info

Schloss Dagstuhl, Germany

Patient records form an important cornerstone of nowadays medical information systems. Technically, they represent complex multimedia documents that comprise temporally interrelated elements of various media types. The collection of patient records of a health care institution forms a digital library, with access limited to this organization. Efficient similarity search in large patient record databases is an important foundation for a number of highly rewarding analysis operations. Therefore we propose a novel approach to similarity retrieval of complex multimedia documents, in general, and patient records, more specifically. That is, we adapt and extend a retrieval algorithm that is known to work efficiently for region matching of images. Specifically, we introduce efficient and effective domain-specific matching of patient records, that (1) exclusively matches elements of identical media types (like x-ray images) with (2) dedicated measures (like nearest neighbor on texture features) and (3) only establishes semantically meaningful matches (like x-rays of same organs in identical orientation). Our approach is not limited to the medical domain and can be used for any large-scale digital library of compound multimedia documents.