NEW APPROACHES TO COMPUTER-ASSISTED DIAGN OSIS IN RHEUMATOLOGY

Since the 1960s, several knowledge-based systems for computer-assisted diagnosis in radiology have been developed. The great majority of the se tools has been implemented as off-line systems. This requires inter action with the system solely for the purpose of consultation and ther efore interrupts the radiologist's work flow. This and inadequate man- machine interfaces may have inhibited the routine clinical use of such systems. The goal of this paper is to describe the current research t oward the development of the online expert system Cadiag-4/Rheuma-Radi o. The underlying fundamentals of the system design, including client/ server architecture, communication interfaces, and fuzzy set theory an d fuzzy logic as methods for knowledge representation and interference , are presented. Methods: In radiology today, computers are routinely used to acquire radiological images in hospital and radiology informat ion systems (HIS/RIS) and picture archiving and communication systems (PACS). In our approach, we make use of pre-existent sources of inform ation to build an expert system that minimizes the interaction between radiologists and the computer. To handle uncertainty and vagueness of medical knowledge, fuzzy set theory and fuzzy logic are used. Given d ata of a specific case, a deductive inference procedure combines the o bserved radiological signs, establishes confirmed and excluded diagnos es as well as diagnostic hypotheses, and provides explanations for the se conclusions. Furthermore, proposals for confirmation or exclusion o f diagnostic hypotheses are offered. Results: For evaluation purposes, an early prototype of Cadiag-4/Rheuma-Radio was tested on radiologica l disorders of the hip joint related to rheumatological diseases. Twen ty radiological cases were used as test cases, reaching a diagnostic a ccuracy of about 80 %. Conclusion: The first results are acceptable an d encourage further work to cover the whole area of rheumatologically relevant radiological signs and diagnoses. Furthermore, research into the development of user-oriented data acquisition tools will be carrie d out.