This document is also available in these non-normative formats: html version.
Copyright 2005, Agfa HealthcareThis note summarizes selected aspects of our investigations into the use of proof-based technologies to validate and manage medical knowledge. In the course of our investigations we have found great benefit in being able to validate, restructure, and extend the data using a variety of tools and proof based engines. As our experience grows in this area, a number of key requirements are emerging. This paper discusses some specific use cases and then re-casts them in terms of requirements on the underlying rule based systems.
This section describes the status of this document at the time of its publication. Other documents may supersede this document. This document is maintained at Agfa Healthcare.
This document has been prepared as a position paper for the W3C Workshop on Rule based systems.
A list of all current W3C related Agfa Healthcare Technical Reports can be found at http://www.agfa.com/w3c/Notes/Overview.html.
1 Introduction
2 Details
2.1 Refactoring
2.2 Expressiveness.
2.3 Rapid Authoring
2.4 Validation.
2.5 Named Formal Definitions.
2.6 Extendability.
2.7 Interoperability.
2.8 Re-Use of Knowledge.
3 Conclusions
The past decade has seen amazing advances in our ability to acquire medical facts. These facts range from new clinical or theoretical discoveries (see [WIHIR]) to our ability to collect and utilize large volumes of patient diagnostic and treatment data. This information has the potential to drastically affect peoples medical treatment and lives in both a positive and negative manner. Proper usage of medical knowledge can literally mean life. Errors of ommission or outright mistakes can mean death or injury.
Our investigations so far have shown that the semantic web is an effective mechanism for representing a wide range of medical knowledge. For example, we have been successful in recasting published clinical guidelines (see [TRIP]) in a usable form using n3 and have been able to use a variety of proof engines to validate clinical contexts against such guidelines.
The sheer volume of largely unstructured, constantly changing facts, procedures and guidelines together with the very human need for assurances of correctness and validation against established procedures make this subject area an ideal area for applications of semantic web technologies.
Our experiments to this point made use of [N3] and rules together with systems such as Euler [EULER], CWM [CWM] and Jena [JENA] to model situations requiring both forward and backward reasoning and take into account some experience with describing the semantics for content MathML [MATHML].
In the course of these investigations we have been able to make several observations concerning the rules, the data, and their interaction.
The key items that we wish to take into consideration when reviewing the infrastructure of rules are outlined below.
We have had a continual need to refactor our rules as we discover new factors that come into play and struggle with finding the right combination of facts and meta-like rules such as transitivity versus collections of expanded rules
There has been a continual tradeoff between expressiveness and the decidability of the resulting queries.
In the early stages of development (at least until visual editing tools mature) the ability to quickly write down meaningful facts and statements in a human readable form has been essential.
The ability to easily author rules in a human understandable format is essential to developments in this area. History has shown that this was essential to early development of the Web. This only becomes less important after usable editors become widely available. Another case in point is MathML. Given the complexity of writing advanced semantic constructs in MathML the successful authoring of MathML is closely tied to the use of tools.
There is a need for easy methods of validation of both syntax and against a chosen structure. In the long term, this may be addressed by authoring tools, but quick convenient validation is essential to successful authoring.
As we work with different implementations we have developed a library of Rules. There would appear to be a need include rules by name.
It is important to allow for systems that support different logical models. How do we advertise the capabilities of a system? How do we behave when desired functionality is not supported?
We need to at least consider mechanisms for naming individual rules. This can have considerable impact on our ability to quickly author and experiment. Conventions such as name spaces in order to be clear when a named rule is really the same rule is essential.
As the collection of proof based technologies matures and different balances between expressivness and decidability are explored, the need to plan for extendability becomes important.
We should not have to re-design the language to add new functionality. There are potential lessons to be learned in the design of content MathML and in its appendices providing more formal mathematical definiions.
We have needed to consult a variety of proof engines and tools during our investigations. while a diversity of features has been benificial, the notions of discoverability and interoperability emerge, and might even benefit from the notion of a compatibilty layer.
During the course of our investigations we have had the constant need to feed results from one system into another. One of the key outcomes of this activity must be to establish that interoperability.
Througout, it has proved helpful to be able to separate the core "re-usable" knowledge from temporal or state specific knowledge and to use meta-structure whereever possible.
The actual data that is produced by a proof must also be valid input to the next proof, or for direct addition to our knowledge pool.
We do not presume to have absolute answers to the questions raised here. Nor do we presume that such a list is complete. An effective design will be the result of balancing these needs and most of all through buy-in by the stake-holders. A satisfactory outcome of such a workshop would be some consensus of where to draw the line and an explanations of how the emerging design addresses each of these questions. We look forward to working with participants to move forward in an organized fashion.