A neural network expert system to support decisions in diagnostic mapping

An expert system module has been integrated into clinical workstations used for report generation and the storage and retrieval of diagnostic images. The expert system works in the background and supports the physician´s decisions. The system has been created for ultrasound examinations of the abdomen. The knowledge-based system has been implemented for the same knowledge segment using two different methods: a rule-based approach with the AI-tool Joshua on a Symbolics machine for prototyping and knowledge debugging and a neural network approach based on back propagation. The results show that the neural network is robust and flexible for the mainly associative knowledge found in this application; it performed well in general and, specifically, on subsets of findings entered during the course of an examination. It was therefore decided to combine both components