Knowledge graph

A structured map of things and the relationships between them: this drug treats this condition, this condition presents with this finding, this finding is seen on this modality. Instead of holding facts as free text, a knowledge graph holds them as entities and links, which a machine can follow step by step and a person can inspect. SNOMED CT, ICD, and the medical ontologies underneath them are close relatives, so most hospitals already run on one without calling it that. The current interest is in pairing a graph with a language model, so the model's fluency gets checked against a source that states what is actually true. When a vendor says its tool reasons over clinical knowledge, the useful question is which graph or ontology it uses and who keeps that source current.

An interaction checker that fires when two drugs on the list conflict is a knowledge graph doing its job, and it is the reason you can see exactly why it fired. Compare that with a language model asked the same question, which returns a fluent answer with nothing behind it you can inspect. The pairing is where this is heading: the model handles the wording, the graph decides what is true. When a tool claims clinical knowledge, ask which source it traverses and how often that source is updated.

Terms like this come up in real clinical scenarios across the HelloAI courses: bite-sized modules with verifiable certificates. An account takes one minute, no password needed.