Active research and competition system

Clinical Entity Recognition and Linking Pipeline

Started
2026-06
Last revised

A Vietnamese clinical NLP pipeline that separates span detection, assertion handling, candidate retrieval, and ontology linking so each source of error can be inspected independently.

  • XLM-R
  • Linear-chain CRF
  • PyTorch
  • BM25
  • Dense retrieval
  • Cross-encoder reranking
  • RxNorm
  • ICD
RQ

Research question

How can a Vietnamese clinical system preserve precise entity boundaries and assertion context while linking noisy surface forms to controlled medical vocabularies with limited annotated data?

01

From input to inspectable output

System architecture

Entity path · deterministic interfaces
01Note textnormalized tokens
02XLM-R + CRFtyped spans
03Assertioncontext state
04ALexicalalias overlap
04BDensesemantic context
05Rerank + linkRxNorm / ICD
Fig. 01Recognition and assertion stay observable before two retrieval paths merge.
  1. 01

    Clinical text normalization

    Preserve medically meaningful punctuation while standardizing whitespace and known abbreviation variants.

    in
    note text
    out
    token sequence
  2. 02

    Span recognition

    XLM-R contextual representations feed a CRF so adjacent BIO decisions are decoded as a valid sequence.

    in
    tokens
    out
    typed spans
  3. 03

    Assertion layer

    A separate contextual classifier assigns present, absent, possible, or historical status without changing span boundaries.

    in
    span + context
    out
    asserted entity
  4. 04A

    Lexical candidate retrieval

    Normalized aliases and character-aware lexical search recover terminology with strong surface overlap.

    in
    entity mention
    out
    top-k lexical IDs
  5. 04B

    Dense candidate retrieval

    Multilingual mention and concept embeddings recover semantic candidates missed by exact terminology.

    in
    entity + context
    out
    top-k semantic IDs
  6. 05

    Candidate fusion and reranking

    Deduplicated candidates are calibrated and reranked with the local sentence before an ontology ID is selected.

    in
    candidate union
    out
    RxNorm / ICD link
02

What the design must respect

Operating constraints

  1. 01

    Clinical abbreviations, spelling variation, and mixed Vietnamese–English terminology make exact lexical matching brittle.

  2. 02

    The annotation budget is limited, so the pipeline must make useful progress before a large task-specific corpus exists.

  3. 03

    Negation, hypothetical mentions, and family-history context must remain attached to an entity after recognition.

  4. 04

    RxNorm and ICD candidate spaces differ in granularity; a shared retrieval score is not automatically comparable.

  5. 05

    Competition-time inference needs bounded candidate sets and reproducible preprocessing.

03

Evaluation ledger

Current results

Placeholder results — no benchmark claim is being made.

Values will be added only with the evaluation set, protocol, ontology version, and relevant runtime conditions.

R01Entity span F1
Not reportedPlaceholder

Reserved for a held-out, versioned evaluation set; no score is claimed yet.

R02Assertion macro F1
Not reportedPlaceholder

Will be split by present, absent, possible, and historical classes.

R03Linking recall@20
Not reportedPlaceholder

The first retrieval measure to report, before reranking accuracy.

R04End-to-end linked entity F1
Not reportedPlaceholder

Requires the span, assertion, and ontology versions to be frozen together.

04

Negative evidence

Failed or insufficient approaches

F01

Single-stage exact alias matching

Useful as a precision-oriented baseline, but it cannot resolve abbreviations, inflections, or context-dependent concepts.

Design response

Keep exact matches as a retrieval feature and add lexical plus dense candidate generation.

F02

Linking every raw model span immediately

Boundary errors become opaque linking errors, which makes evaluation difficult to interpret.

Design response

Evaluate span detection, candidate recall, and reranking separately before measuring the end-to-end path.

F03

One similarity threshold for every ontology

Scores are not calibrated across concept families or between RxNorm and ICD inventories.

Design response

Calibrate by ontology and entity type, and retain an explicit abstention option.

F04

Dropping assertion context after NER

A correctly linked medication can still be clinically misleading when the note says it was discontinued or denied.

Design response

Treat assertion as part of the entity record and carry it through serialization and evaluation.

05

What changed in the model of the problem

Lessons

  • Candidate recall is the ceiling for a reranker; retrieval diagnostics should come before model complexity.
  • Span-boundary and ontology-link errors need separate labels in the error ledger.
  • Character-aware lexical signals remain valuable beside multilingual embeddings.
  • Abstention is a meaningful prediction when the terminology does not support a defensible link.
  • Dataset, ontology, and normalization versions belong in every experiment record.
06

Queue, not promises

Next experiments

  1. 01Compare reciprocal-rank fusion with a learned lexical–dense score calibration layer.
  2. 02Measure candidate recall by entity type and by normalized mention frequency.
  3. 03Test whether assertion-aware context improves linking for medication and diagnosis mentions.
  4. 04Build a small adjudicated challenge slice containing abbreviations and overlapping spans.
  5. 05Profile the accuracy–latency trade-off for cross-encoder reranking at different candidate depths.
07
  1. Practical Error Analysis for Biomedical Named Entity Recognition

    A taxonomy for turning span errors into tractable experiments.

  2. Dense Retrieval Versus Lexical Retrieval for Medical Entity Linking

    Where the two candidate generators fail, and why a hybrid remains useful.

  3. Understanding Viterbi Decoding in Linear-Chain CRFs

    The dynamic program behind constrained sequence decoding.