A PSYCHIATRIC DIAGNOSTIC SYSTEM INTEGRATING PROBABILISTIC AND CATEGORICAL REASONING

We describe a diagnostic support system for clinical psychiatry and it s evaluation results. The system has two inter-related components: a r ule-based reasoning part associated with uncertainty, and a determinis tic part, that uses heuristics to perform categorical reasoning. The s ystem includes the 30 groups of psychiatric diagnoses which are classi fied under the categories 290 to 319 of the DSM-III-R and the ICD-9. T here are, in fact, 1508 rules relating 208 clinical findings with 257 diagnoses. The reasoning strategy is based on selecting and differenti ating diagnostic categories in a hierarchical classification tree. The system is intended to be used for education of medical students, and to help non-specialist clinicians, residents in psychiatry, or experts with few years of experience in decision making. We tested the diagno stic performance of the system using case reports extracted from a spe cialized journal. in 52.8% of the cases, the correct diagnosis was ran ked as the first hypothesis using only the rule-based part. In combina tion with the deterministic strategy, the correct diagnosis could be m ade for 73.6% of the analyzed cases.