health-safe-ocr 0.1.0

Safe OCR pipeline for healthcare images with PII masking and NHS number validation.

health-safe-ocr

Safe OCR pipeline for healthcare images with PII masking, NHS number validation, confidence scoring, and human review flags.

health-safe-ocr helps developers extract text from healthcare images before sending content to an LLM, RAG pipeline, vector database, logs, audit system, or internal API.

Installation

pip install health-safe-ocr

For local development:

python -m venv .venv
.venv\Scripts\activate
pip install -e ".[dev]"

On macOS or Linux:

python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev]"

Tesseract OCR Requirement

This package uses pytesseract, which requires the Tesseract OCR system binary to be installed separately.

Windows:

winget install UB-Mannheim.TesseractOCR

macOS:

brew install tesseract

Linux:

sudo apt install tesseract-ocr

Python Usage

from health_safe_ocr import HealthSafeOCR

ocr = HealthSafeOCR()
result = ocr.extract(
    "patient_letter.jpg",
    mask_sensitive=True,
)

print(result.text)
print(result.masked_text)
print(result.confidence)
print(result.requires_human_review)
print(result.entities)

Example result:

OcrResult(
    text="Patient Sachin Tiwari NHS No 485 777 3456",
    masked_text="Patient Sachin Tiwari NHS No [NHS_NUMBER]",
    confidence=0.88,
    requires_human_review=False,
    entities=["NHS_NUMBER"],
    engine="tesseract",
)

CLI Usage

Extract plain text:

health-safe-ocr patient_letter.jpg

Extract and mask PII:

health-safe-ocr patient_letter.jpg --mask-pii

Return JSON:

health-safe-ocr patient_letter.jpg --mask-pii --json

Example JSON output:

{
  "text": "Patient Sachin Tiwari DOB 12/05/1990 NHS No 943 476 5919",
  "masked_text": "Patient Sachin Tiwari [DOB] NHS No [NHS_NUMBER]",
  "confidence": 0.88,
  "requires_human_review": false,
  "entities": ["DOB", "NHS_NUMBER"],
  "engine": "tesseract"
}

Supported PII Masking

Version 0.1.0 supports regex-based masking for:

• Email addresses
• UK phone numbers
• Valid NHS numbers
• Dates of birth
• UK postcodes
• National Insurance numbers
• Patient IDs

Safety Note

This package does not provide diagnosis, clinical advice or treatment recommendations. It only extracts text from images and masks sensitive information.

Always review OCR output before using it in clinical, operational, audit, or production AI systems. OCR confidence scores and regex masking can miss sensitive data.

Local Checks

Run tests:

pytest

Run the CLI locally:

health-safe-ocr examples/sample.jpg --mask-pii --json

Build package distributions:

python -m build

Upload to TestPyPI:

python -m twine upload --repository testpypi dist/*

Upload to PyPI:

python -m twine upload dist/*

Roadmap

0.1.0

• Image OCR
• PII masking
• NHS number validation
• CLI
• JSON output
• PyPI release

0.2.0

• PDF support
• Scanned PDF page extraction
• Better image preprocessing

0.3.0

• PaddleOCR engine option
• Better OCR accuracy
• Multilingual support

0.4.0

• Microsoft Presidio integration
• Advanced entity detection
• Name and address masking

0.5.0

• Azure Document Intelligence provider
• Cloud OCR fallback

1.0.0

• Stable API
• Full docs
• Production examples
• FastAPI integration
• RAG-safe pipeline
• Audit report output

License

MIT