Bỏ qua tới nội dung chính
Clinic Hub Logo
Clinic Hub
Blog
← Quay lại Blog
AI in clinic operations

Digitize paper medical records with AI (OCR): a 5-step process

Clinic Hub
digitizationocrmedical recordsdata privacy

Đọc bài này bằng tiếng Việt →

Cabinets full of visit notes, lab results and referral letters are a familiar sight in many private clinics. When you digitize paper medical records, information is faster to find and less likely to go missing. Text recognition (OCR), now often assisted by AI, can turn a photo of a visit note into text. But OCR is not magic: it needs good images and a person to check the result. Here is a 5-step process you can apply right away, even without any software.

Where paper records hurt

  • Slow searches: digging through cabinets by name or date while the patient waits.
  • Lost or misfiled sheets: a lab result clipped into the wrong file is as good as lost.
  • Hard-to-read handwriting, especially years later or when someone else wrote it.
  • Only one copy: damp or a flood and it is gone for good.

What OCR can and cannot do

OCR (optical character recognition) reads the text in an image and turns it into text you can copy and search. On clearly printed paper, such as a machine-printed lab result, the output is often usable.

But know the limits:

  • Handwriting, especially hurried handwriting, is easily misread or skipped.
  • Numbers, units, decimal points and Vietnamese diacritics are easy to get wrong, and they are exactly what matters most in a medical record.
  • Tables with several columns can come out with rows mixed up.
  • OCR copies text; it does not understand medical content. WHO also warns that large multi-modal AI models can produce false or incomplete statements.

So OCR text is a draft for lookup. The original image is the reference.

The 5-step process

Step 1: Sort and prioritize records

Do not try to scan the whole cabinet in a week. Start with:

  • patients with upcoming appointments or on long-term treatment;
  • patients with chronic conditions that need follow-up;
  • papers you look up often: lab results, referral letters, allergy information.

Scan only as much per session as you have time to check.

Step 2: Photograph or scan properly

Image quality decides most of the OCR quality.

  • Even light, no hand shadows, no flash glare.
  • Flat paper, with the camera parallel to the page.
  • Capture the whole page, without cutting off margins or signatures.
  • Sharp enough that the smallest text is still readable when you zoom in.
  • One page per image; if there are notes on the back, photograph the back too.
  • No beauty or colour filters.
  • Check each image straight away and retake it if it is blurry.

Step 3: Name and link each file to the right patient

A file linked to the wrong patient is more dangerous than a lost file.

  • Match at least two identifiers (name and year of birth, or patient code) before linking.
  • Use one naming convention: patient code, document type, date on the paper.
  • Categorize: lab result, X-ray, referral, other.

Step 4: Run OCR and review the text

Put the OCR text next to the original image and read them side by side. Check first: name, year of birth and visit date; numbers, units and reference ranges; recorded medicine names and allergy information. Mark anything you cannot read as unreadable; do not guess. Unreviewed text should not be used for clinical decisions.

Step 5: Keep the original and control access

  • Keep the original image file next to the OCR text; when in doubt, always open the original.
  • Do not destroy paper records after scanning without legal advice on retention periods.
  • One account per staff member; only people who need documents can see them.
  • Do not send record photos through personal messaging apps.

A note on privacy

Decree 356/2025/NĐ-CP lists health status and images of ID cards as sensitive personal data. It requires organizations processing this data to set access-restriction rules, processing procedures and security measures. Photos of visit records usually contain health information, and sometimes an ID document too.

Know where images are processed: on the clinic's own machine, or sent to a cloud AI service. If it is the cloud, ask the vendor where processing happens, how long images are kept and what is logged. This is not legal advice; ask a lawyer before processing real patient data.

Vietnam's Ministry of Health (Bộ Y tế) also has separate rules for electronic medical records (Circular 13/2025/TT-BYT, in effect since 21 July 2025), under which records are created, signed and stored electronically. Scanned images with OCR text for lookup do not automatically become an electronic medical record under those rules.

How Clinic Hub helps

  • A Documents tab in every patient record. Upload photos or PDFs, categorize them (lab result, X-ray and more), link them to the patient and, optionally, to a visit slip.
  • OCR text next to the original file. Uploaded documents join an OCR queue with a clear status. The "OCR result" view shows the document and the recognized text side by side for checking, with a "Re-run OCR" button. The original file stays in the record.
  • Two ways to run OCR. An OCR worker running on a clinic Mac recognizes text on the machine itself, and images are not sent to any AI provider. Cloud AI vision OCR is off by default and is switched on per clinic. It reads JPEG, PNG and WebP images (PDF and HEIC need the on-device worker) and only picks up recently uploaded documents on its own; it does not send your whole document archive.
  • Scan paperwork into a registration draft. Photograph a patient's paperwork and cloud AI prefills name, year of birth, gender, phone, email and address in the patient form. Staff check every field and any duplicate warning before saving; closing the dialog saves nothing.
  • Logging and permissions. Every cloud AI scan is recorded in the AI log. The clinic owner grants access page by page for each member.

Want to try this workflow? Request access or browse the feature list.

Sources

Related posts