Using schemas for diagnosis

Medical diagnosis can be considered a planning task: actions are decided on, then executed. However, unlike traditional planning tasks, new information can arise any time, changing the situation and invalidating the plan. In order to cope with this, the reasoner must be able to interleave planning with plan execution. Our approach to this problem is called schema-based reasoning. In this approach, the reasoner's procedural knowledge is represented as schemas, small packets of procedural knowledge which can be used to achieve a goal. When a new goal arises, the reasoner uses the goal and the features of the current consultation to retrieve an appropriate schema from memory; the schema is then applied to achieve the goal. If several schemas are active, the reasoner chooses between them using information about the current situation and information from strategic schemas. Our approach is implemented in MEDIC, a schema-based diagnostic reasoner whose domain is pulmonology.