Similarity Search Applications in Medical Images

The advances in medical imaging have enabled the large-scale generation of medical image data in hospitals and private clinics. While taking a high-resolution scan of a troubling body part has been rather exceptional in the past, it is now common practise for a large variety of patient conditions.

The research project THESEUS MEDICO aims at the re-organization of medical picture archiving and communication systems (PACS) such that they can be efficiently queried for more advanced image meta-data than a small set of common identifiers. As a complement to keyword-based retrieval queries, MEDICO also offers a variety of content-based image retrieval (CBIR) applications.

Two use cases employing CBIR techniques for answering clinical queries on Computed Tomography (CT) scans have been developed in the course of this thesis. The first application provides a standardized body coordinate system using image-based landmarks and further image information for automatically providing an anatomical context for any given scan region and for greatly accelerating retrieval queries. The second application is a more direct form of CBIR, enabling the specification of a visual query template which is used for retrieving similar image patterns.

This thesis introduces a variety of newly-developed similarity search components and it evaluates the generated medical search framework at the example of the two search applications.