Guide
Translating medical documents: a practical guide for healthcare professionals
Why medical translation is different
Generic translation tools fail medical documents not because the language is difficult — but because the consequences of an error are unusually high. A mistranslated drug name, a wrong dosage unit, an ambiguous consent clause — each is a patient safety issue, not just a quality issue.
Medical translation requires three things that most tools don't provide together: terminology precision, layout fidelity, and an auditable quality process.
Terminology precision: the critical constraint
Medical terminology is highly standardised. ICD-10 codes, drug generic names (INN), anatomical nomenclature, and procedure codes (CPT) all have agreed international forms. A translation that paraphrases rather than uses the standard term introduces ambiguity and can cause downstream errors.
Traxlate handles medical terminology in two ways:
Glossary enforcement. Before translation, you can define a term list: drug names, procedure codes, anatomical terms, and institution-specific vocabulary. These terms are pinned before translation runs and are guaranteed to appear verbatim in the output. This is not optional for clinical documents; it is the only way to guarantee term consistency.
Editorial polish and accuracy verification. Every translation is produced and polished as a first draft, then independently checked against the source — anything that drifts is flagged for review. The polish stage operates at the discourse level: it corrects pronoun resolution across paragraphs, enforces formal register, and catches contextually plausible but medically incorrect phrasing. For consent forms and clinical protocols, add a human polish pass on top.
OCR for scanned records
Most medical records are not born-digital. Paper charts, faxed referrals, discharge summaries printed from legacy EHR systems — all of these arrive as image files or scanned PDFs.
The OCR pipeline matters enormously for these documents. Traxlate handles the full range:
1. Digital text extraction for born-digital PDFs: clean, correctly ordered text without any image processing.
2. Balanced OCR for standard scans: language-aware recognition with denoising and auto-deskew. Appropriate for clean scans at 200 DPI or better.
3. High-quality OCR for degraded documents: a slower, more thorough pass that significantly improves accuracy on faded, low-contrast, or handwritten material. Adds 30–60 seconds.
4. Auto mode: balanced first, escalates automatically to high-quality if confidence is weak. Recommended for most healthcare workflows — fast for clean documents, safe for degraded ones.
Layout preservation
Discharge summaries, operative notes, and lab reports have specific layout requirements. Field labels must align with values. Table columns must be preserved. Headers must be distinct from body text.
Traxlate's document reconstruction stage uses the bounding box of each detected text block to reflow translated text into the same spatial positions. Tables are identified and preserved as data regions — each cell is translated independently and reflowed into the original table structure.
The result looks like the original document with translated text, not a flat text dump.
When to use human review
For documents that inform treatment decisions, or documents that carry legal force (consent forms, insurance pre-authorisations, regulatory submissions), machine translation is a first draft, not a final product.
Traxlate's human polish flow works as follows:
1. The machine translation run completes.
2. You request a human polish pass from the job detail page.
3. A professional linguist with appropriate domain expertise reviews and corrects the machine output.
4. The polished translation is delivered to your dashboard (24–72 hours depending on language pair and document length).
The cost is 500 credits per 1,000 source characters, with a language-pair uplift for rare pairs. For clinical documentation, the human polish pass is the appropriate level of assurance.
Data handling and jurisdiction
Medical documents contain PHI (protected health information) under HIPAA and personal health data under GDPR. Jurisdiction matters.
Traxlate is GDPR-compliant end-to-end, with no CLOUD Act exposure on the processing path. We offer:
- Configurable data retention (7 days to 1 year; immediate deletion available)
- Per-key IP allowlists for API access
- EU DPA (data processing agreement) on request
- No training on user documents
For enterprise healthcare customers, contact support@traxlate.com for compliance documentation.
Practical workflow example
A radiology department receives image reports from partner clinics in French and German and needs English summaries for US-based specialists.
1. Upload the scanned PDF from the PACS system
2. Set OCR profile to auto, source to French or German, target to English
3. Pin radiology terminology (e.g. "malignant" not "cancerous", "hypointense" not "dark") via the glossary manager
4. Download the translated report as a DOCX with the original layout
5. For reports that will be used in treatment planning, add a human polish pass
Total time for a 4-page radiology report: 60–90 seconds machine translation, 24–48 hours if human polish is requested.