Automated Interpretation of Diagrams for Specification of Medical Protocols
John W. Egar, Ángel R. Puerta, Mark A. Musen · 1992
Automated Interpretation of Diagrams for Specification of Medical ProtocolsJohn W. Egar, Angel R. Puerta, Mark A. Musen.Knowledge Systems LaboratoryStanford University, Stanford, CA 94305-5479 USAWe are building a visual programming environment that enables medical researchers to organize and de-scribe complex research protocols (specified treatment plans) so that this information can be incorporatedinto decision-support systems. We have designed our system with a five-layer architecture and a collectionof compilers for various visual languages. This design permits the interchange of modules and the coordi-nation of different visual languages within the same system.1. ObjectiveAlthough diagramming environments are often excellent tools for modeling complex relationships, they are also dif-ficult to implement. Our research team is building knowledge-based tools that will help researchers to design and executemedical treatment plans [1, 2]. One of our tasks in this project is to create several visual programming languages that clin-ical researchers and computer scientists can use to model the problem-solving methods and the medical background in-formation for many protocols in a particular field, and the structure of individual protocols. This paper describes how wehave implemented these visual languages.2. Design ConsiderationsPrevious experience with domain-specific knowledge-acquisition tools [2, 3] has convinced us that flowcharts are aparticularly useful way for expert physicians to describe the management of clinical trials for experimental therapies. Weare now including classification trees, state charts, entity-relationship diagrams, influence diagrams, and data-flow dia-grams to describe the necessary domain information and problem-solving methods [1].Environments for multiple visual programming languages do exist [4, 5, 6, 7, 8, 9]. With few exceptions [4, 10, 11,12], these environments have relied to some degree on syntax-directed editing. These editors guide the user so that sheconstructs only syntactically correct diagrams. This strategy requires a custom-tailored editor for each visual language. Inits extreme form, syntax-directed editing forces the user to model in a top-down fashion, introducing successively less ab-stract placeholders according to rigid replacement rules. Because of these limitations, we have elected instead to treat di-agrams as syntax-neutral documents that are compiled after they are saved.Certain interface dynamics are also necessary for such a diagramming tool to be successful. The ease and speed ofuse of the tool is of particular concern. The ability to nest diagrams within nodes and edges, and thereby to avoid muchvisual clutter, is another criterion. Hypertext links to text and forms are desirable, as is the ability to interchange notationalconventions to suit the expert rather than the system. Other design goals include reliability, n~ainminability, portability, andmodularity. Finally, a simple graphical user-interface convention, such as drag-and-drop, is a valuable feature.3. System DescriptionThe implementation of such a battery of visual languages presents a formidable programming task. After building twocustom-tailored diagram editors, we developed the following five-layered approach:186