Incremental and interaction-based knowledge acquisition for medical images in THESEUS

Abstract

Today, the major challenge in medical imaging is the so called knowledge acquisition bottleneck. We cannot acquire the necessary medical image knowledge that ought to be used in the software application easily as it is hidden in the heads of medical experts. In this article, we provide an example of how an incremental knowledge acquisition process for radiology images can be implemented to solve this problem. Thereby, we integrated Semantic Web technologies with a variety of automatic and manual annotation tools for radiology images. We developed the prototypes in the context of a large scale German research program for a new Internet infrastructure based on semantic technologies - THESEUS. According to the complex medical finding processes in the MEDICO use case, the different annotation tools should be used for very specific purposes. After four years of prototyping automatic and manual annotation tools (2009-2012), we developed a divide-and-conquer strategy for future knowledge acquisition processes. This divide-and-conquer strategy turns out to be very effective in the radiology domain, but produces many infrastructure requirements. It also relies on high-end intelligent user interfaces such as the Radspeech dialogue system which are not available in today's clinical environments.