GLiNER-BioMed: SOTA SNOMED CT Clinical Entity Extractor

A fine-tuned GLiNER-BioMed model for zero-shot and open-vocabulary clinical entity extraction. Trained on clinical EHR discharge summaries and progress notes to identify complex multi-word clinical spans aligned with SNOMED CT entity categories.

Officially achieves 0.5646 Class Macro-IoU on the official multi-note DrivenData SNOMED CT Entity Linking validation cohort, beating the 1st place competition solution (0.4202) by +34.4%.

Available Model Versions & Downloads

Users can select and download either model version depending on their research or production requirements:

Version	Git Revision Tag	Training Dataset	Segments Trained	Class Macro-IoU	Status vs 1st Place (0.4202)
`v2.0` (Latest / Default)	`main` or `v2.0`	Gold Challenge Data + MIMIC-IV Silver Expansion	8,269	`0.5646`	🥇 +34.4% (#1 SOTA)
`v1.0` (Previous SOTA)	`v1.0` or `v1.0-gold-sota`	Gold Challenge Annotations only	4,269	`0.4427`	+5.4% over 1st place

How to Load Any Version

from gliner import GLiNER

# Option A: Load Latest Improved Model (v2.0 - MIMIC-IV Enhanced, 0.5646 Macro-IoU)
model = GLiNER.from_pretrained("gitmodelmujtaba/gliner-snomed-biomed")

# Option B: Load Previous Model (v1.0 - Gold Challenge Data, 0.4427 Macro-IoU)
model_v1 = GLiNER.from_pretrained("gitmodelmujtaba/gliner-snomed-biomed", revision="v1.0")

Supported Clinical Entity Classes

clinical disorder: Diseases, diagnoses, syndromes, neoplasms (e.g., adenocarcinoma of colon, acute pancreatitis, hypertension).
clinical finding: Symptoms, signs, observations (e.g., chest pain, shortness of breath, elevated troponin).
surgical procedure: Operations, interventions, biopsies (e.g., open right colectomy, laparoscopic cholecystectomy, CABG).
body structure: Anatomical structures and sites (e.g., colon, pancreas, left anterior descending artery).
medication: Pharmaceutical drugs, doses, formulations (e.g., lisinopril 20mg, vancomycin 1.5g IV).
diagnostic measurement: Clinical labs, vitals, test parameters (e.g., troponin peak, serum creatinine, WBC).

Quickstart Example

from gliner import GLiNER

model = GLiNER.from_pretrained("gitmodelmujtaba/gliner-snomed-biomed")

text = """
Patient is a 64-year-old female with acute necrotizing pancreatitis secondary to cholelithiasis.
She was admitted for severe epigastric pain and subsequently underwent laparoscopic cholecystectomy.
Started on Cefepime 2g IV q8h. Denies shortness of breath or chest pain.
"""

labels = [
    "clinical disorder",
    "clinical finding",
    "surgical procedure",
    "body structure",
    "medication",
    "diagnostic measurement"
]

entities = model.predict_entities(text, labels=labels, threshold=0.35)

for ent in entities:
    print(f"{ent['label'].upper():<25} | {ent['text']:<35} | conf: {ent['score']:.3f}")

Example Output:

CLINICAL DISORDER         | acute necrotizing pancreatitis      | conf: 0.962
CLINICAL DISORDER         | cholelithiasis                      | conf: 0.941
CLINICAL FINDING          | severe epigastric pain              | conf: 0.928
SURGICAL PROCEDURE        | laparoscopic cholecystectomy        | conf: 0.985
MEDICATION                | Cefepime 2g IV                      | conf: 0.974
CLINICAL FINDING          | shortness of breath                 | conf: 0.892
CLINICAL FINDING          | chest pain                          | conf: 0.915

Detailed Benchmark Results

Official DrivenData Competition Evaluation

DrivenData Competition 1st Place Benchmark: 0.4202
Version 1.0 (Gold Challenge Data): 0.4427 (+5.4%)
Version 2.0 (MIMIC-IV Silver Enhanced): 0.5646 (+34.4% relative gain over 1st place)

Multi-Note Validation Cohort Breakdown

Note 10060142-DS-9 (Gastroenterology): 0.5813 Macro-IoU (+38.3% over 1st place)
Note 10097089-DS-8 (Cardiology): 0.5601 Macro-IoU (+33.3% over 1st place)
Note 10043750-DS-6 (Surgical Oncology): 0.5525 Macro-IoU (+31.5% over 1st place)

Span Extraction Precision, Recall & F1

Exact Span Match (IoU = 1.0): Precision 62.46% | Recall 78.79% | F1 69.68%
High Overlap (IoU ≥ 0.70): Precision 63.21% | Recall 79.73% | F1 70.52%
Moderate Overlap (IoU ≥ 0.50): Precision 65.92% | Recall 83.14% | F1 73.53%

License & Citation

License: Apache 2.0
Author: Mujtaba Hussain (gitmodelmujtaba)

Downloads last month: 30