⚡ STEAM Clinical Assertion

Find your clinical terms in a note and classify each mention's assertion status.

How it works — two steps

① Find the terms — word-anchored matching (~O(n)). Every term is located in a single linear pass over the note. Tokens are clinical-aware, so 1.5, p.o., mg/dl, and covid-19 stay intact and a term only matches whole tokens.

② Classify — STEAM BERT. Each located mention is then classified by the BERT model on five assertion axes (negated · certainty · realis · experiencer · temporality), which roll up to a single i2b2 assertion label.

i2b2 labels:
present absent possible hypothetical associated with someone else
Load a sample note (optional)

Classified terms