BIDS standardizes file layout. It does not standardize what your phenotypic variables
mean. One dataset records dx_group, another diagnosis, a third patient_status — and
nothing tells a machine these are the same concept, or that “MCI” here matches “mild
cognitive impairment” there.
Semantic annotation closes that gap by mapping each variable onto a controlled vocabulary such as SNOMED CT. Neurobagel is the Canadian-led implementation: annotate a dataset’s variables, publish the resulting graph, and datasets become searchable by participant characteristics across institutions — without moving the underlying data.
This is directly relevant to anyone here pooling cohorts, which is a premise of several CNS projects.
Related¶
Nipoppy — already mentions Neurobagel
Semantic embedding for clinical metrics — an LLM-based approach to a related problem