On Limitations of Using Rough Set Approach to Analyse Non-trivial Medical Information Systems

The rough set theory has been used to analyse medical experience with urolithiasis patients treated by extracorporeal shock wave lithotripsy (ESWL). The aim of this analysis was to evaluate the signi cance of attributes for two classi cations expressing the patients' condition after the ESWL treatment and to discover strong decision rules representing classi cation patterns interesting for practitioners. The ESWL information system is an example of non-trivial medical data set where the use of a simple rough set model gives a high number of possible reducts which are impossible to interpret. Two heuristic strategies based on the rough set theory are proposed. They lead to the selection of the most signi cant attributes having a good clinical interpretation. Inducing only discriminating decision rules do not give good interpretation results rules are weak and too speci c. Discovery of partly discriminating rules allows to extract strong classi cation patterns.