AsbruView: Visualization of Time-Oriented, Skeletal Plans

Skeletal plans are a powerful way to reuse existing domainspecific procedural knowledge. The main drawbacks are that the compositions and the interdependencies of different skeletal plans and their components are not lucid. The aim of this paper is to overcome these limitations and to present the visualization of time-oriented, skeletal plans. Within the Asgaard project, we have developed a time-oriented and intention-based language, called Asbru, to represent such skeletal plans. The Asbru syntax is defined in Backus-Naur form (BNF). Reading BNF or similar forms are next to impossible even for domain experts. We explored different representations and automated knowledge-acquisition tools. However, the domain experts did not accept any of these representations. Consequently, we investigated different metaphor graphics and ended up with a plan visualization utilizing the metaphors of "tracks" and "traffic", called AsbruView. We formatively evaluated different approaches of this plan visualization with physicians applying treatment protocols of mechanical ventilated newborn infants. Introduction and Motivation Our approach is oriented, but not limited to our application domain: the medical (high-frequency) domain. Physicians are faced with two problems: (1) the information overload resulting from modern equipment, and (2) improving the quality of health care through increased awareness of proper disease management techniques. Treatment planning from scratch typically is not necessary, as general procedures exist which should guide the medical staff. These procedures are called clinical guidelines or protocols. Appropriate clinical protocols are only available for a very limited class of clinical problems. Mostly, they are expressed in natural language, but this kind of representation can not easily be transformed into a formal and structured framework (Herbert 1994). Additionally, clinical protocols are not adjusted to the patient data-management system and they are partly vague and incomplete concerning their intentions and their temporal, context-dependent representation. Clinical protocols are a way of pre-compiling decisions that must be made, in which experts’ ° Copyright © 1997, American Association for Artificial Intelligence (www.aaai.org). All rights reserved. knowledge is distilled into a form of procedural knowledge. Extracting and formulating the knowledge structure for protocols is a non trivial task. Their implicit context must be made explicit. The variability of clinical protocols presents an additional challenge. A medical goal can be achieved by different therapeutic actions (e.g., in the domain of mechanical ventilation: pressure-controlled, volume-controlled, or ratio-controlled ventilation). Besides using natural language, the most favored attempts to capture and support clinical protocols, are flow diagrams and flowcharting tools. Many medical experts are used to working with these techniques. Flow diagrams are useful for representing sequential states and actions in a graphical way. However, it is quite difficult to cope with all possible orders of plan execution and all the exception conditions that might arise. The trouble is that this by necessity can only cover a small subset of the possible situations and possible paths through. Additionally, flow diagrams are kinds of layering, which avoid to cope with concurrent (parallel) actions, with different temporal dimensions, with high numbers of possible transitions, and with mutual dependencies of parameters. A possible way to overcome these limitations are skeletal plans. Similar problems were solved by procedural reasoning systems (PRSs, Georgeff, Lanskey, and Schoppers 1986) and situated and reactive planning (Firby 1989; Suchman 1987; Wilkins and Myers 1995). However, we need to have greater temporal reasoning power and focus on issues such as temporally extended goals (Bacchus and Kabanza 1996).