Medical magnetic resonance imaging (MRI) is a new expensive imaging modality which avoids ionizing radiation and provides excellent diagnostic images. The images produced depend on both the properties of the patient's tissues and the radiologist controlled parameters of the MRI scanner. When scanner parameters are selected with clearcut objectives accounting for the patients's history, anatomic relationships and the MR properties of relevant tissues, accurate diagnostic images result. There are expert MRI radiologists who understand these tradeoffs, and use heuristic reasoning to simplify the mathematical and clinical complexity. Their expertise can be encoded into a knowledge-based expert computer system. Specifically, we propose to accomplish the following research: 1. Using existing methods of artificial intelligence and knowledge engineering, construct a first prototype expert system. It will be designed to incorporate the following functionality: o Be a useful consultant for twenty representative situations involving the anatomy of the head, such as multiple sclerorsis, acoustic neuroma and pituitary adenoma. o In a given clinical setting, produce a set of MR imaging parameters optimal for determining if a specific disease process is present. o Be able to explain its reasoning in terms comprehensible to radiologists. 2. Explore different inference strategies for reasoning about complex MRI physics interactions, such as constraint algorithms and mixed forward/backward chaining approaches. 3. Build a user interface uniquely suited to the needs of radiologists engaged in MR imaging. 4. Expand the knowledge of the prototype by incorporating changes in MR imaging technology and gradually including more anatomical regions. 5. Develop automated knowledge acquisition strategies which help maintain the accuracy and completeness of the knowledge base. This represents new research in artificial intelligence.
