Tetra: A Case-Based Decision Support System for Assisting Nuclear Physicians with Image Interpretation

摘要

This paper shows how nuclear image interpretation is improved by TETRA, a case-based decision support system. TETRA exploits two kinds of knowledge sources: ontologies and knowledge embedded in past nuclear imaging reports, each imaging report being associated with a case, described by some features and its associated diagnoses. Ontologies are used, in addition with vocabulary resources, to semantically annotate the imaging reports. Links between case features and diagnoses in the training case base have been computed. In practice, when a new image test is run, TETRA exploits this features/diagnosis links, as well as the generalization/specialization relation of the ontologies to retrieve the cases that are the most similar to the new image test and to compute the most probable diagnoses. 8000 nuclear imaging reports have been collected to create a case base and almost 1000 other imaging reports have been used for the system evaluation, which shows that TETRA gives good results for the two diagnoses (necrosis and ischemia) which have been considered in this work. The first results shows that an ontology-based similarity computation between cases in order to display the most similar cases as well as the diagnosis probability computation helps the nuclear physician in her image interpretation task.