openmed 1.2.0

OpenMed delivers state-of-the-art biomedical and clinical LLMs that rival proprietary enterprise stacks, unifying model discovery, advanced extractions, and one-line orchestration.

OpenMed

Production-ready medical NLP toolkit powered by state-of-the-art transformers

Transform clinical text into structured insights with a single line of code. OpenMed delivers enterprise-grade entity extraction, assertion detection, and medical reasoning—no vendor lock-in, no compromise on accuracy.

from openmed import analyze_text

result = analyze_text(
    "Patient started on imatinib for chronic myeloid leukemia.",
    model_name="disease_detection_superclinical"
)

for entity in result.entities:
    print(f"{entity.label:<12} {entity.text:<35} {entity.confidence:.2f}")
# DISEASE      chronic myeloid leukemia            0.98
# DRUG         imatinib                            0.95

---

✨ Why OpenMed?

• Specialized Models: 12+ curated medical NER models outperforming proprietary solutions
• HIPAA-Compliant PII Detection: Smart de-identification with all 18 Safe Harbor identifiers
• One-Line Deployment: From prototype to production in minutes
• Dockerized REST API: FastAPI endpoints for service deployments
• Batch Processing: Multi-file workflows with progress tracking
• Production-Ready: Configuration profiles, profiling tools, and medical-aware tokenization
• Zero Lock-In: Apache 2.0 licensed, runs on your infrastructure

---

Quick Start

Installation

# From a local checkout of this repo:
# Install with Hugging Face support
uv pip install -e ".[hf]"

# Or include REST service dependencies
uv pip install -e ".[hf,service]"

Apple Silicon acceleration in Python:

uv pip install -e ".[mlx]"

Swift apps on macOS and iOS use OpenMedKit. In 1.2.0, that means:

• MLX on Apple Silicon macOS and real iPhone/iPad hardware for supported OpenMed PII, OpenAI Privacy Filter, and experimental GLiNER-family artifacts
• CoreML when you already have a bundled Apple model package or want the fallback Apple path

Add the Swift package like this:

dependencies: [
    .package(url: "https://github.com/maziyarpanahi/openmed.git", from: "1.2.0"),
]

OpenMedKit is public and now supports native MLX runtime paths for PII token classification, Privacy Filter, and experimental GLiNER-family zero-shot tasks. The broader OpenMed model-packaging flow is still being hardened across the full collection, so treat conversion as active work rather than a fully universal public release surface.

For published releases, the editable install examples above can be replaced with plain uv pip install "openmed[...]".

Three Ways to Use OpenMed

1️⃣ Python API — One-liner for scripts and notebooks

from openmed import analyze_text

result = analyze_text(
    "Patient received 75mg clopidogrel for NSTEMI.",
    model_name="pharma_detection_superclinical"
)

2️⃣ REST API Service — FastAPI endpoints for app backends

uvicorn openmed.service.app:app --host 0.0.0.0 --port 8080

3️⃣ Batch Processing — Programmatic multi-document workflows

from openmed import BatchProcessor

processor = BatchProcessor(
    model_name="disease_detection_superclinical",
    confidence_threshold=0.55,
    group_entities=True,
)

result = processor.process_texts([
    "Patient started metformin for type 2 diabetes.",
    "Imatinib started for chronic myeloid leukemia.",
])

---

Key Features

Core Capabilities

• Curated Model Registry: Metadata-rich catalog with 12+ specialized medical NER models
• PII Detection & De-identification: HIPAA-compliant de-identification with smart entity merging
• Medical-Aware Tokenization: Clean handling of clinical patterns (COVID-19, CAR-T, IL-6)
• Advanced NER Processing: Confidence filtering, entity grouping, and span alignment
• Multiple Output Formats: Dict, JSON, HTML, CSV for any downstream system

Production Tools (v1.2.0)

• Batch Processing: Multi-text and multi-file workflows with progress tracking
• Configuration Profiles: dev/prod/test/fast presets with flexible overrides
• Performance Profiling: Built-in inference timing and bottleneck analysis
• Dockerized REST API: GET /health, POST /analyze, POST /pii/extract, POST /pii/deidentify
• Service Reliability Hardening: request validation, shared pipeline preload, and timeout/error envelopes

---

Documentation

Comprehensive guides available at openmed.life/docs

Quick links:

• Getting Started — Installation and first analysis
• Analyze Text Helper — Python API reference
• MLX Backend — Apple Silicon Python runtime
• OpenMedKit (Swift) — Use OpenMed models in macOS and iOS apps
• PII Detection Guide — Complete de-identification tutorial (v0.5.0)
• Batch Processing — Multi-text and multi-file workflows
• Configuration Profiles — Environment-specific presets
• REST Service — FastAPI and Docker usage
• Model Registry — Browse available models
• Configuration — Settings and environment variables

---

REST API

OpenMed includes a Docker-friendly FastAPI service with reliability hardening:

• GET /health
• POST /analyze
• POST /pii/extract
• POST /pii/deidentify

Run locally

uv pip install -e ".[hf,service]"
uvicorn openmed.service.app:app --host 0.0.0.0 --port 8080

Optional shared model warm-up:

OPENMED_SERVICE_PRELOAD_MODELS=disease_detection_superclinical,OpenMed/OpenMed-PII-SuperClinical-Small-44M-v1 \
uvicorn openmed.service.app:app --host 0.0.0.0 --port 8080

Run with Docker

docker build -t openmed:1.2.0 .
docker run --rm -p 8080:8080 -e OPENMED_PROFILE=prod openmed:1.2.0

Example request

curl -X POST http://127.0.0.1:8080/pii/extract \
  -H "Content-Type: application/json" \
  -d '{"text":"Paciente: Maria Garcia, DNI: 12345678Z","lang":"es"}'

See the full service guide at REST Service docs.

Non-2xx responses now use a unified envelope:

{
  "error": {
    "code": "validation_error",
    "message": "Request validation failed",
    "details": [
      {
        "field": "body.text",
        "message": "Text must not be blank",
        "type": "value_error"
      }
    ]
  }
}

---

Models

OpenMed includes a curated registry of 12+ specialized medical NER models:

Model	Specialization	Entity Types	Size
disease_detection_superclinical	Disease & Conditions	DISEASE, CONDITION, DIAGNOSIS	434M
pharma_detection_superclinical	Drugs & Medications	DRUG, MEDICATION, TREATMENT	434M
pii_detection_superclinical	PII & De-identification	NAME, DATE, SSN, PHONE, EMAIL, ADDRESS	434M
anatomy_detection_electramed	Anatomy & Body Parts	ANATOMY, ORGAN, BODY_PART	109M
gene_detection_genecorpus	Genes & Proteins	GENE, PROTEIN	109M

📖 Full Model Catalog

---

Advanced Usage

PII Detection & De-identification (v0.5.0)

from openmed import extract_pii, deidentify

# Extract PII entities with smart merging (default)
result = extract_pii(
    "Patient: John Doe, DOB: 01/15/1970, SSN: 123-45-6789",
    model_name="pii_detection_superclinical",
    use_smart_merging=True  # Prevents entity fragmentation
)

# De-identify with multiple methods
masked = deidentify(text, method="mask")        # [NAME], [DATE]
removed = deidentify(text, method="remove")     # Complete removal
replaced = deidentify(text, method="replace")   # Synthetic data
hashed = deidentify(text, method="hash")        # Cryptographic hashing
shifted = deidentify(text, method="shift_dates", date_shift_days=180)

Smart Entity Merging (NEW in v0.5.0): Fixes tokenization fragmentation by merging split entities like dates (01/15/1970 instead of 01 + /15/1970), ensuring production-ready de-identification.

HIPAA Compliance: Covers all 18 Safe Harbor identifiers with configurable confidence thresholds.

📓 Complete PII Notebook | 📖 Documentation

Multilingual PII (9 Languages)

OpenMed now supports multilingual PII extraction and de-identification across en, fr, de, it, es, nl, hi, te, and pt. French, German, Italian, and Spanish expose the full 35-model family; Portuguese ships 31 public API-visible models; Dutch, Hindi, and Telugu currently ship one flagship public model each, bringing the total PII catalog to 210 models.

uv pip install "openmed[hf]" && python -c "from openmed import extract_pii; print([(e.label,e.text) for e in extract_pii('Dr. Pedro Almeida, CPF: 123.456.789-09, email: [e-mail], tel: [phone]', lang='pt').entities])"

from openmed import extract_pii

portuguese = extract_pii(
    "Paciente: Pedro Almeida, CPF: 123.456.789-09, email: [e-mail], telefone: [phone]",
    lang="pt",
    model_name="OpenMed/OpenMed-PII-Portuguese-SnowflakeMed-Large-568M-v1",
    use_smart_merging=True,
)

dutch = extract_pii(
    "Patiënt: Eva de Vries, geboortedatum: 15 januari 1984, BSN: 123456782, telefoon: [phone]",
    lang="nl",
    model_name="OpenMed/OpenMed-PII-Dutch-SuperClinical-Large-434M-v1",
    use_smart_merging=True,
)

hindi = extract_pii(
    "रोगी: अनीता शर्मा, जन्मतिथि: 15 जनवरी 1984, फोन: [phone], पता: 12 गली संख्या 5, नई दिल्ली 110001",
    lang="hi",
    model_name="OpenMed/OpenMed-PII-Hindi-SuperClinical-Large-434M-v1",
    use_smart_merging=True,
)

telugu = extract_pii(
    "రోగి: సితా రెడ్డి, జన్మ తేదీ: 15 జనవరి 1984, ఫోన్: [phone], చిరునామా: 12 వీధి 5, హైదరాబాద్ 500001",
    lang="te",
    model_name="OpenMed/OpenMed-PII-Telugu-SuperClinical-Large-434M-v1",
    use_smart_merging=True,
)

print([(e.label, e.text) for e in portuguese.entities])
print([(e.label, e.text) for e in dutch.entities])
print([(e.label, e.text) for e in hindi.entities])
print([(e.label, e.text) for e in telugu.entities])

Batch Processing

from openmed import BatchProcessor, OpenMedConfig

config = OpenMedConfig.from_profile("prod")
processor = BatchProcessor(
    model_name="disease_detection_superclinical",
    config=config,
    group_entities=True,
)

result = processor.process_texts([
    "Metastatic breast cancer treated with trastuzumab.",
    "Acute lymphoblastic leukemia diagnosed.",
])

Configuration Profiles

from openmed import analyze_text

# Apply a profile programmatically
result = analyze_text(
    text,
    model_name="disease_detection_superclinical",
    config_profile="prod"  # High confidence, grouped entities
)

Performance Profiling

from openmed import analyze_text, profile_inference

with profile_inference() as profiler:
    result = analyze_text(text, model_name="disease_detection_superclinical")

print(profiler.summary())  # Inference time, bottlenecks, recommendations

📖 More Examples

---

Contributing

We welcome contributions! Whether it's bug reports, feature requests, or pull requests.

• 🐛 Found a bug? Open an issue

---

License

OpenMed is released under the Apache-2.0 License.

---

Citation

If you use OpenMed in your research, please cite:

@misc{panahi2025openmedneropensourcedomainadapted,
      title={OpenMed NER: Open-Source, Domain-Adapted State-of-the-Art Transformers for Biomedical NER Across 12 Public Datasets},
      author={Maziyar Panahi},
      year={2025},
      eprint={2508.01630},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2508.01630},
}

---

Star History

If you find OpenMed useful, consider giving it a star ⭐ to help others discover it!

---

Built with ❤️ by the OpenMed team

🌐 Website • 📚 Documentation • 🐦 X/Twitter • 💬 LinkedIn