genzeonplatform/cliniguard-ner

CliniGuard NER -- PHI/PII De-identification by Genzeon Platform

CliniGuard NER is a clinical Named Entity Recognition model developed by Genzeon Platform for automated detection and de-identification of Protected Health Information (PHI) and Personally Identifiable Information (PII) in clinical text. Built on a domain-specialized BERT architecture fine-tuned on healthcare corpora, CliniGuard delivers production-grade entity recognition across 20 PHI categories.

Model Details

Property	Value
Developed by	Genzeon Platform
Architecture	BertForTokenClassification
Parameters	~110M
Tagging scheme	BIO (41 labels)
Max sequence length	512 tokens
License	Apache-2.0

Intended Use

CliniGuard NER is designed for enterprise healthcare environments where patient data privacy is critical. Primary use cases include:

• Clinical text de-identification -- removing or masking patient identifiers before sharing medical records for research.
• PII detection -- flagging sensitive information in healthcare documents, EHRs, and discharge summaries.
• Regulatory compliance -- supporting HIPAA Safe Harbor de-identification requirements.
• Healthcare AI pipelines -- preprocessing clinical text for downstream NLP tasks while ensuring patient privacy.

Entity Types

The model recognizes 20 PHI entity types using BIO tagging (41 labels total):

Category	Entity Types
Patient identifiers	PATIENT_NAME, DATE_OF_BIRTH, AGE, GENDER, SSN, MRN
Contact information	PHONE, FAX, EMAIL
Location	ADDRESS, CITY, STATE, ZIP, COUNTRY
Organization	HOSPITAL
Provider	DOCTOR_NAME
Digital identifiers	USERNAME, ID_NUMBER, IP_ADDRESS, URL

Performance

Overall Metrics

Metric	Precision	Recall	F1
Micro avg	0.9659	0.9732	0.9695
Macro avg	0.9609	0.9706	0.9656

Per-Entity Metrics

Entity	Precision	Recall	F1	Support
PATIENT_NAME	0.9817	0.9853	0.9835	14335
DATE_OF_BIRTH	0.9798	0.9740	0.9769	9818
AGE	0.9028	0.9854	0.9423	1508
GENDER	0.9596	0.9885	0.9738	1562
SSN	0.9513	0.9935	0.9719	766
MRN	0.9938	0.9923	0.9930	1943
PHONE	0.9730	0.9869	0.9799	2590
FAX	0.9481	0.9454	0.9468	696
EMAIL	0.9965	0.9936	0.9950	4543
ADDRESS	0.9746	0.9844	0.9794	1985
CITY	0.9086	0.8891	0.8988	2047
STATE	0.9103	0.9060	0.9082	2734
ZIP	0.9770	0.9832	0.9801	951
COUNTRY	0.9485	0.9504	0.9495	2056
HOSPITAL	0.9033	0.9345	0.9186	5267
DOCTOR_NAME	0.9865	1.0000	0.9932	802
USERNAME	0.9689	0.9431	0.9559	1917
ID_NUMBER	0.9724	0.9898	0.9811	8555
IP_ADDRESS	0.9892	0.9924	0.9908	926
URL	0.9910	0.9947	0.9928	3001

Usage

from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline

model_name = "genzeonplatform/cliniguard-ner"

# Option 1: Use the transformers pipeline
nlp = pipeline("token-classification", model=model_name, aggregation_strategy="simple")
text = "Patient John Smith, DOB 03/15/1960, was seen at Springfield General Hospital by Dr. Jane Doe."
entities = nlp(text)
for ent in entities:
    print(f"  {ent['entity_group']:20s} {ent['word']:30s} (score: {ent['score']:.3f})")

# Option 2: Manual inference
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForTokenClassification.from_pretrained(model_name)

import torch
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=512)
with torch.no_grad():
    outputs = model(**inputs)
predictions = torch.argmax(outputs.logits, dim=2)
tokens = tokenizer.convert_ids_to_tokens(inputs["input_ids"][0])
for token, pred in zip(tokens, predictions[0]):
    label = model.config.id2label[str(pred.item())]
    if label != "O":
        print(f"  {token:20s} -> {label}")

Training Details

• Developed by: Genzeon Platform
• Architecture: Domain-specialized BERT fine-tuned on clinical corpora
• Training data: Genzeon Platform's proprietary clinical NER dataset with diverse healthcare note formats
• Epochs: 15 (with early stopping, patience=3)
• Learning rate: 3e-5 (linear schedule with warmup)
• Batch size: 16 (train) / 32 (eval)
• Max sequence length: 512 tokens
• Optimizer: AdamW (weight decay 0.01)

Limitations

• English only: Currently optimized for English clinical text. Multilingual support is on the Genzeon Platform roadmap.
• Recommended with human-in-the-loop: For high-stakes de-identification workflows, Genzeon Platform recommends pairing CliniGuard with human review for maximum safety.
• Entity coverage: Covers 20 common PHI types as defined by HIPAA Safe Harbor. Rare or domain-specific identifiers may require custom fine-tuning -- contact Genzeon Platform for enterprise support.
• Context window: Limited to 512 tokens per input. Longer documents should be chunked with overlap for best results.

About Genzeon Platform

Genzeon Platform is a healthcare technology company specializing in AI-powered solutions for clinical data management, regulatory compliance, and healthcare interoperability. CliniGuard NER is part of Genzeon Platform's suite of healthcare AI tools designed to accelerate clinical research while safeguarding patient privacy.

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