> ## Documentation Index
> Fetch the complete documentation index at: https://docs.nometria.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Medical OCR

> Multi-engine OCR pipeline for medical and legal documents — extracts ICD codes, CPT codes, medications, and structured data.

`medical-ocr` is a multi-engine OCR pipeline for medical and legal documents. It extracts structured data — ICD codes, CPT codes, medications, timelines, impairment ratings — from PDFs and scanned documents.

<CardGroup cols={2}>
  <Card title="GitHub" icon="github" href="https://github.com/nometria/medical-ocr">
    nometria/medical-ocr
  </Card>

  <Card title="PyPI" icon="python" href="https://pypi.org/project/medical-ocr/">
    medical-ocr on PyPI
  </Card>
</CardGroup>

## Install

```bash theme={null}
# System dependencies
brew install tesseract poppler          # macOS
apt-get install tesseract-ocr poppler-utils  # Ubuntu

# Install base package
pip install medical-ocr

# With GPU-accelerated OCR (EasyOCR + OpenCV)
pip install medical-ocr[gpu]

# With Google Cloud Vision fallback
pip install medical-ocr[gcp]
```

## Usage

```bash theme={null}
# Set API key for LLM refinement pass
export OPENAI_API_KEY=sk-proj-...

# Process a medical document — extract all fields
medical-ocr report.pdf --all --format json

# Extract specific fields only
medical-ocr report.pdf --fields icd,cpt,medications

# Batch process a directory
medical-ocr ./patient-records/ --all --format json --output ./extracted/

# Run as REST API
medical-ocr --api
# POST /extract  { "file": "path/to/document.pdf" }
```

## Extraction pipeline

| Step              | Engine              | What it does                        |
| ----------------- | ------------------- | ----------------------------------- |
| **OCR**           | Tesseract (primary) | Extracts raw text from pages        |
| **Secondary OCR** | EasyOCR (optional)  | Higher accuracy for handwriting     |
| **Fallback OCR**  | Google Cloud Vision | For complex layouts                 |
| **Classify**      | Rules + LLM         | Identifies document type            |
| **Extract**       | LLM refinement      | Pulls structured data from OCR text |

## Supported document types

| Document type       | Extracted fields                                     |
| ------------------- | ---------------------------------------------------- |
| Treatment records   | Diagnoses, procedures, dates, providers              |
| Prescriptions       | Medications, dosages, frequencies, refills           |
| Imaging reports     | Body part, findings, impression, radiologist         |
| IME reports         | Impairment ratings, restrictions, causation opinions |
| Operative reports   | Procedure codes (CPT), surgeons, facility            |
| Discharge summaries | Admission/discharge dates, diagnoses, follow-up      |
| Lab results         | Test names, values, reference ranges, flags          |
| Bills/EOBs          | CPT codes, charges, allowed amounts, dates           |

## Output format

```json theme={null}
{
  "document_type": "treatment_record",
  "icd_codes": ["M54.5", "G89.29"],
  "cpt_codes": ["99213", "97110"],
  "medications": [
    { "name": "Ibuprofen", "dosage": "800mg", "frequency": "TID" }
  ],
  "body_parts": ["lumbar spine", "right shoulder"],
  "work_restrictions": ["no lifting > 10 lbs", "limited standing"],
  "dates": {
    "first_visit": "2025-01-10",
    "last_visit": "2025-03-15"
  }
}
```