SBIR-STTR Award

The need to apply expert knowledge in the interpretation of medical imagery and sensor data is well documented. Rule-based expert systems have proven their value in such applications as medical expertise. Most diagnostic systems mimic the consulting-room environment, acquiring symptom descriptions and lab findings through dialog with the user. Automated diagnosis from imagery and analog is a more difficult problem. The proposed effort seeks to advance the field by addressing knowledge acquisition problems in applying expert systems technology to imagery and signal data. This problem can be addressed by an expert system that writes rule bases. Similar to a knowledge engineer, a rule-writing system learns what the expert knows. We propose a program of research to develop a system called metarule that will advance the state-of-the-art in knowledge based systems-specifically, in the areas of machine learning and knowledge representation; lead to the development of a commercially viable inductive reasoning system incorporating the advances required to deal with imagery and signals; and enable expert systems to be developed that would otherwise be infeasible including non-signals expert systems.

The application of expert systems technology to the interpretation of telemetry data, such as medical imagery and sensor data, offers substantial payoffs. Less experienced physicians would be able to perform as well as experts; expertise would be conserved when an expert departs; and diagnoses would be made consistent and reproducible. But the knowledge underlying expert telemetry interpretation is complex, involving vision and pattern recognition as well as rules. The proposed effort covers the development of a workstation which will use advanced pattern recognition hardware and software together with machine learning to learn and apply a physician's diagnostic criteria for medical telemetry. Feasibility was demonstrated under a prior Phase I effort. The workstation will develop rule bases for particular diagnostic problems. A diagnostician can select and run a rule base to obtain advice and assistance in the diagnosis of a case. the first application will be the diagnosis of planar thallium myocardial imagery. At the end of the effort, the workstation will be installed at the school of aerospace medicine, Brooks AFB.