SEMANTIC MODELING OF DISEASE AND TREATMENT IN AI-BASED MEDICAL SYSTEMS

Abstract. Artificial intelligence in healthcare is often discussed as an algorithmic problem, yet its reliability depends just as strongly on how clinical meaning is represented. This mini-review examines how semantic modeling structures disease and treatment information for AI-based medical systems. A focused narrative synthesis of eight peer-reviewed publications published between 2004 and 2024 was conducted across four domains: terminology integration, logic-based clinical vocabularies, FHIR-based semantic structures, and EHR-derived knowledge graphs. The reviewed literature suggests that dependable medical AI requires more than data aggregation. It requires normalized concepts, typed relations, contextual qualifiers, and traceable provenance. From this synthesis, a four-layer schema emerged: concept normalization, relation modeling, clinical context, and governance. Systems built on these layers are more likely to support transportable prediction, explainable reasoning, and safer clinical deployment. Future work should extend semantic models beyond diagnosis coding toward richer treatment representations that capture intent, sequence, dosage, response, and uncertainty.