A knowledge-based system for biomedical image processing and recognition

The purpose of this paper is to describe a rule-based system named IBIS (Interpretation of Biomedical Images of the Slice type), which, starting from conventional image processing, can recognize slices and locate automatically their principal anatomical organs in nuclear magnetic resonance (NMR) images of the head. The main aspects of the system are: knowledge representation, with 3-D and 2-D aspects for anatomical description; error recovery and conflict resolution facilities at different backtracking levels; and a general system structure easy to extend to other application fields. Production rules are employed in order to define the criteria for deciding what action is to be performed on the basis of the current problem status. However, in order to obtain a more explicit and appropriate representation, anatomical knowledge has not been spread throughout the production rules, but it has been incorporated into an extended semantic network that well represents relational properties. The control structure consults production rules and, according to their content, activates the various processing procedures in order to attain the recognition goal. A strategy is applied that can be driven by data or models, depending on the processing phase and on progressive results. Applications of the system to different patients and slices are described and discussed. The importance of the paper lies basically in the methodological approach to image processing and interpretation, and in possible interesting applications to the medical field.