EVALUATION OF AUTOMATIC KNOWLEDGE ACQUISITION TECHNIQUES IN THE DIAGNOSIS OF ACUTE ABDOMINAL-PAIN

Clinical diagnosis in acute abdominal pain is still a major problem. C omputer-aided diagnosis offers some help; however, existing systems st ill produce high error rates. We therefore tested machine learning tec hniques in order to improve standard statistical systems. The investig ation was based on a prospective clinical database with 1254 cases, 46 diagnostic parameters and 15 diagnoses. Independence Bayes and the au tomatic rule induction techniques ID3, NewId, PRISM, CN2, C4.5 and ITR ULE were trained with 839 cases and separately tested on 415 cases. No major differences in overall accuracy were observed (43-48%), except for NewId, which was below the average. Between the different techniqu es some similarities were found, but also considerable differences wit h respect to specific diagnoses. Machine learning techniques did not i mprove the results of the standard model Independence Bayes. Problem d imensionality, sample size and model complexity are major factors infl uencing diagnostic accuracy in computer-aided diagnosis of acute abdom inal pain.