Upload README.md with huggingface_hub

c204b8a verified 6 days ago

17 kB

language:
  - en
license: apache-2.0
library_name: transformers
tags:
  - transformers
  - safetensors
  - bert
  - ner
  - clinical
  - medical
  - healthcare
  - biomedical
  - pubmedbert
  - named-entity-recognition
  - token-classification
  - diagnosis
  - icd-10
  - icd-coding
  - snomed
  - sapbert
  - entity-linking
  - disease
  - ehr
  - hipaa
datasets:
  - synthetic-clinical-diagnosis
  - mimic-iii
  - ncbi-disease
  - share-clef
  - i2b2-2010
pipeline_tag: token-classification
model-index:
  - name: healthcare-brain-diagnosis-icd-ner
    results:
      - task:
          type: token-classification
          name: Named Entity Recognition
        metrics:
          - type: f1
            value: 0.9347
            name: F1 (Strict)
          - type: precision
            value: 0.9341
            name: Precision
          - type: recall
            value: 0.9353
            name: Recall

Healthcare Brain Diagnosis ICD NER — Diagnosis & ICD Coding Entity Extraction by Genzeon Platform

Healthcare Brain Diagnosis ICD NER is a transformer-based clinical Named Entity Recognition model developed by Genzeon Platforms for automated extraction of diagnoses, conditions, and support for ICD-10/SNOMED code mapping from unstructured clinical text. Built on PubMedBERT and fine-tuned on clinical diagnosis corpora, this model delivers production-grade entity recognition across 9 diagnosis and coding entity categories.

The model implements a two-stage pipeline: (1) NER extraction of diagnosis mentions, (2) Entity linking using SapBERT for automated ICD-10/SNOMED code mapping.

Model Details

Property	Value
Developed by	Genzeon Platforms
Base model	PubMedBERT (microsoft/BiomedNLP-PubMedBERT-base-uncased-abstract-fulltext)
Architecture	BERT Token Classification (BIO tagging) + SapBERT Entity Linking
Parameters	~110M (NER) + ~110M (SapBERT linker)
Tagging scheme	BIO (19 labels)
Max sequence length	512 tokens
Framework	HuggingFace Transformers
License	Apache-2.0

Intended Use

Healthcare Brain Diagnosis ICD NER is designed for healthcare AI pipelines that need to extract structured diagnosis information from unstructured clinical text. Primary use cases include:

Diagnosis extraction — extracting primary, secondary, differential, and comorbid diagnoses from discharge summaries, progress notes, and clinical narratives.
ICD-10/SNOMED coding support — automated mapping of extracted diagnosis mentions to standardized medical codes using SapBERT entity linking.
Clinical documentation improvement (CDI) — identifying diagnosis specificity gaps for coding accuracy and DRG optimization.
Complication detection — identifying hospital-acquired complications and comorbidities for quality reporting.
Clinical research — extracting diagnosis-related entities from large clinical corpora for epidemiological and outcomes studies.

Entity Types

The model recognizes 9 diagnosis and coding entity types using BIO tagging (19 labels total):

Category	Entity Type	Description	Examples
Primary	`PRIMARY_DIAGNOSIS`	Principal reason for encounter	acute myocardial infarction, pneumonia
Secondary	`SECONDARY_DIAGNOSIS`	Additional diagnoses	type 2 diabetes, hypertension
Differential	`DIFFERENTIAL_DIAGNOSIS`	Diagnoses under consideration	pulmonary embolism vs pneumonia
Comorbidity	`COMORBIDITY`	Co-existing conditions	hyperlipidemia, obesity
Complication	`COMPLICATION`	Hospital-acquired or treatment complications	acute kidney injury, sepsis
Chronic	`CHRONIC_CONDITION`	Long-term conditions	COPD, coronary artery disease
Acute	`ACUTE_CONDITION`	Acute episodes or events	acute respiratory failure, DKA
Status	`DIAGNOSIS_STATUS`	Current clinical status	active, resolved, improving
Temporal	`DIAGNOSIS_DATE`	Date of diagnosis or onset	03/15/2024, on admission, 2 weeks ago

Note: External dataset loaders (MIMIC-III, NCBI Disease, ShARe/CLEF, i2b2 2010 Problems) are architecturally supported and included in this release. These datasets require Data Use Agreements from PhysioNet, i2b2.org, and CLEF respectively. Contact Genzeon Platforms for enterprise models trained with full real-world clinical data coverage.

Performance

Overall Metrics

Metric	Precision	Recall	F1
Micro avg	0.9341	0.9353	0.9347
Macro avg	0.9286	0.9240	0.9262

Per-Entity Metrics (Strict: Exact Span + Exact Type)

Entity	Precision	Recall	F1	Support
DIAGNOSIS_STATUS	0.9523	0.9571	0.9547	903
PRIMARY_DIAGNOSIS	0.9487	0.9513	0.9500	757
COMORBIDITY	0.9434	0.9494	0.9464	908
DIAGNOSIS_DATE	0.9401	0.9418	0.9410	653
CHRONIC_CONDITION	0.9365	0.9397	0.9381	614
COMPLICATION	0.9241	0.9189	0.9215	407
ACUTE_CONDITION	0.9189	0.9073	0.9131	259
SECONDARY_DIAGNOSIS	0.9056	0.8898	0.8976	118
DIFFERENTIAL_DIAGNOSIS	0.8873	0.8609	0.8739	115

Two-Stage Pipeline: NER + Entity Linking

Stage 1: Named Entity Recognition

PubMedBERT-based token classification extracts diagnosis mentions from clinical text.

Stage 2: SapBERT Entity Linking

Extracted mentions are encoded using SapBERT and matched to ICD-10/SNOMED concept embeddings via nearest-neighbor lookup.

from src.inference.predictor import DiagnosisEntityLinker

linker = DiagnosisEntityLinker(
    model_dir="genzeonplatform/healthcare-brain-diagnosis-icd-ner",
    code_index_path="path/to/icd10_index.json",
)

text = "Patient admitted with acute myocardial infarction. PMH: type 2 diabetes."
results = linker.link(text, top_k=3)
for r in results:
    print(f"{r['diagnosis']} -> {r['linked_codes'][0]['code']} ({r['linked_codes'][0]['description']})")

Usage

from transformers import pipeline

# Load the model
nlp = pipeline(
    "token-classification",
    model="genzeonplatform/healthcare-brain-diagnosis-icd-ner",
    aggregation_strategy="simple",
)

# Process clinical text
text = \"\"\"Discharge Diagnosis: Primary: acute myocardial infarction.
Secondary: type 2 diabetes mellitus, essential hypertension.
Status: improving. Complication: acute kidney injury.\"\"\"

entities = nlp(text)
for ent in entities:
    print(f"  [{ent['entity_group']:25s}] {ent['word']} (score: {ent['score']:.3f})")

Output:

  [PRIMARY_DIAGNOSIS        ] acute myocardial infarction (score: 0.951)
  [SECONDARY_DIAGNOSIS      ] type 2 diabetes mellitus (score: 0.912)
  [COMORBIDITY              ] essential hypertension (score: 0.944)
  [DIAGNOSIS_STATUS         ] improving (score: 0.957)
  [COMPLICATION             ] acute kidney injury (score: 0.929)

Batch Processing

from transformers import pipeline

nlp = pipeline(
    "token-classification",
    model="genzeonplatform/healthcare-brain-diagnosis-icd-ner",
    aggregation_strategy="simple",
)

clinical_notes = [
    "Admitting Diagnosis: community-acquired pneumonia. PMH: COPD, CHF.",
    "Assessment: sepsis is worsening. Complication: acute kidney injury.",
    "CDI query: Principal Dx congestive heart failure with acute respiratory failure.",
    "Differential: pulmonary embolism vs pneumonia. Status: suspected.",
]

for note in clinical_notes:
    entities = nlp(note)
    print(f"Text: {note[:70]}...")
    for ent in entities:
        print(f"  [{ent['entity_group']:23s}] {ent['word']}")
    print()

Structured Output

from transformers import pipeline
import json

nlp = pipeline(
    "token-classification",
    model="genzeonplatform/healthcare-brain-diagnosis-icd-ner",
    aggregation_strategy="simple",
)

text = "Patient with congestive heart failure and acute kidney injury. Status: worsening. Date: 03/15/2024."
entities = nlp(text)

# Structured extraction
structured = [
    {
        "text": ent["word"],
        "type": ent["entity_group"],
        "score": round(ent["score"], 4),
        "start": ent["start"],
        "end": ent["end"],
    }
    for ent in entities
]

print(json.dumps(structured, indent=2))

Training Details

Developed by: Genzeon Platforms
Base model: PubMedBERT (domain-specialized BERT, pre-trained on PubMed abstracts and full-text articles)
NER architecture: BertForTokenClassification (768 → 19 linear head)
Entity linking: SapBERT (cambridgeltl/SapBERT-from-PubMedBERT-fulltext) for ICD-10/SNOMED mapping
Training data: Synthetic clinical diagnosis corpus + NCBI Disease
Epochs: 15 (early stopping, patience=3)
Learning rate: 3e-5 (linear schedule with warmup, 10% warmup ratio)
Batch size: 16 (train) / 32 (eval)
Optimizer: AdamW (weight decay 0.01, gradient clipping 1.0)
Max sequence length: 512 tokens
Best model selection: By entity-level F1 score
Seed: 42

Training Data

Dataset	Split	Samples	Source
Synthetic Clinical Diagnosis	train/dev/test	8,000 / 1,000 / 1,000	Template-based generation (120+ clinical templates)
MIMIC-III Discharge Summaries	train/test	—	PhysioNet (Credentialed DUA required)
NCBI Disease Corpus	train/dev/test	793	NCBI (freely available)
ShARe/CLEF eHealth	train/test	—	CLEF (Organizational DUA required)
i2b2 2010 Problems	train/test	—	i2b2.org (DUA required)

Entity mapping: MIMIC-III ICD codes provide weak supervision for diagnosis extraction. NCBI Disease entities map to PRIMARY_DIAGNOSIS. ShARe/CLEF disorder mentions map to diagnosis categories. i2b2 2010 problem entities are mapped to target categories using clinical context.

Limitations

English only: Currently optimized for English clinical and biomedical text. Multilingual support is on the Genzeon Platforms roadmap.
Synthetic training bias: Primarily trained on template-generated data. Performance on highly variable real-world clinical documentation may differ — contact Genzeon Platforms for enterprise models fine-tuned with restricted clinical datasets (MIMIC, i2b2, ShARe/CLEF).
Diagnosis categorization: Distinguishing PRIMARY_DIAGNOSIS from SECONDARY_DIAGNOSIS and COMORBIDITY depends on contextual cues in the clinical narrative. Ambiguous documentation may affect categorization accuracy.
Entity linking coverage: SapBERT-based ICD-10/SNOMED linking requires a pre-built code index. Coverage depends on the completeness of the index.
Temporal expressions: DIAGNOSIS_DATE extraction handles common clinical date formats but may miss non-standard temporal references.
Human-in-the-loop recommended: For clinical decision-making, coding compliance, and patient safety workflows, pair model predictions with expert clinician or coder review.

Related Genzeon Platforms Models

Healthcare Brain NER — PHI/PII detection and de-identification. 20 PHI categories.
Healthcare Brain Clinical Findings NER — Clinical findings, diseases, conditions extraction. 8 categories.
Healthcare Brain Medication NER — Medication names, dosages, routes, frequencies. 12 categories.
Healthcare Brain Diagnosis NER — Diagnosis extraction with ICD-10/SNOMED linking. 9 categories.
Healthcare Brain Laboratory NER — Laboratory test results, values, units, reference ranges. 10 categories.
Healthcare Brain Vitals NER — Vital signs, body measurements, physiological parameters. 15 categories.
Healthcare Brain Clinical Findings NER — Transformer-based clinical NER model for extraction of clinical findings, diseases, conditions, anatomical locations, and clinical modifiers from clinical text. 8 clinical finding categories, F1: 0.6209 (strict) / 0.968 (relaxed).
Healthcare Brain Medication NER — Transformer-based clinical NER model for extraction of medication names, dosages, routes, frequencies, and administration details from clinical text. 12 medication categories, F1: 0.9272.

About Genzeon Platforms

Genzeon Platforms is a healthcare technology company that is building the agentic AI decision infrastructure for healthcare. The company builds the Healthcare Brain — three production platforms (HIP One, PES One, CPS One) on a patented multi-agent substrate called Aether One™.

Production Deployment

Genzeon Platforms is a participant in the CMS WISeR Innovation Model (2026–2031), operating Medicare FFS prior authorization in New Jersey under MAC JL via Novitas Solutions. Live since January 1, 2026.

Q1 2026 production results:

15k+ cases processed
100% three-day TAT compliance
Zero auto-denials (every non-affirmation signed by a named licensed clinician)
42% reviewer productivity gain
Sub-three-minute median decision latency
85% portal channel adoption

Scale

50+ payer and provider clients across the Genzeon Platforms
1M+ Medicare FFS members served under WISeR

Patent Portfolio

12 USPTO provisional applications filed covering the Aether One™ architecture
Coverage: multi-agent orchestration, atomic criteria decomposition, knowledge containment, dual-channel pharmacy benefit prior authorization, agentic knowledge pack specification, ambient agent integration, and related primitives
~346 claims locked at provisional priority dates
USPTO portfolio anchor #226167

Compliance Posture

SOC 2 Type II
HIPAA compliant
Operates inside the customer perimeter
Supports on-premises, sovereign-cloud, and air-gapped deployments via the Knowledge Containment Architecture (KCA) reference design

Partnerships

10-year Microsoft partnership (5 partner designations, Microsoft Healthcare Agent Service integration, Dragon Copilot extension)
UiPath Platinum (Top 3 HLS)
Available on:
- Azure Marketplace
- AWS Marketplace
- Google Cloud Marketplace
- Salesforce AppExchange

Open Specifications

Genzeon Platforms publishes the Aether Knowledge Pack Specification (AKPS). AKPS enables healthcare coverage policies to be authored as structured markdown that is directly consumable as LLM prompt context.

See: github.com/genzeon/aether-akps

Model Policy

Genzeon Platforms builds on US- and EU-origin open-weight foundation models only (Llama, Gemma, Mistral families) for healthcare and federal deployment contexts. No Chinese-origin models are used in production, position papers, or patent dependent claims.

Headquarters

Exton, Pennsylvania, USA

Genzeon Platforms is a Genzeon company.

Where to Find More

Resource	Link
Company website	https://genzeon.one
Healthcare Brain overview	https://genzeon.one/healthcare-brain
HIP One (clinical reasoning / prior auth)	https://genzeon.one/hip-one
PES One (patient & member engagement)	https://genzeon.one/pes-one
CPS One (AI governance & compliance)	https://genzeon.one/cps-one
Aether One™ architecture	https://genzeon.one/aether-one
Patents	https://genzeon.one/patents
WISeR production deployment	https://genzeon.one/wiser
AKPS open spec	https://github.com/genzeon/aether-akps
Security & trust	https://genzeon.one/security
LinkedIn	https://www.linkedin.com/company/117124252
Contact	https://genzeon.one/contact

Citation

If you use this model or reference Genzeon Platforms in academic, regulatory, or industry work, please cite:

Genzeon Platforms (2026). Healthcare Brain Diagnosis ICD NER is part of Genzeon Platform's suite of healthcare AI tools designed to accelerate clinical research and improve patient care.

For enterprise licensing, custom fine-tuning, or integration support, contact hi@genzeon.one.