SNOMED CT Entity Linking Benchmark

A benchmark for linking text in medical notes to entities in SNOMED Clinical Taxonomy. #health

Benchmark
Open
180 joined

Model: SNO-PI: SNOMED Pre-trained Inference
Published

TeamTeam Pi

Abstract

Linking clinical mentions to SNOMED CT is a zero- and few-shot problem over a very large label space: the 272 annotated MIMIC-IV discharge summaries cover only 6,595 of the roughly 219,000 eligible concepts, and a third of those occur only once. This motivates a design built on transfer from pre-trained models and human-curated knowledge. SNO-PI combines high-recall span proposal, SapBERT candidate retrieval and listwise ensemble reranking with three BERT-family encoders, each further pre-trained with masked language modelling on unlabelled MIMIC-IV notes and then on SNOMED CT synonyms and hierarchy before task fine-tuning. We find that curated knowledge helps most on concepts never seen in training, yet these concepts remain the main source of error.

Technical paper:

SNO-PI: Label-Scarce SNOMED CT Entity Linking with Pre-trained Encoders and Curated Terminology Knowledge (https://arxiv.org/abs/2610.26425).

Submissions (3)

Submissions
Name Macro char IoUSupport-weighted char IoU
Trial Qwen3 8B single pass as reranker
3w 2d ago
0.5387 0.6629
Upgraded rerankers
4w 1d ago
0.5352 0.6575
Retrieval and Ranker
1mo ago
0.5000 0.6276