Anonymization

OCR detect text overlays

Detect burned-in text overlays using OCR and propose redaction boxes.

Problem

DICOM images often contain burned-in text overlays — patient names, hospital IDs, scan dates — rendered directly into the pixel data by the acquisition scanner. Metadata-only anonymization doesn’t touch these pixel-level identifiers. You need to detect and flag these text overlays so they can be redacted from the image before sharing.

Steps

  1. Open your DICOM file (⌘O) and switch to Anon mode (⌘2). The Cornerstone viewport shows the image.

  2. Trigger OCR detection — click the ⌕ Detect text button in the Anonymization panel. The button text changes to “Scanning…” while the OCR engine processes the current image frame.

  3. Wait for scanning to complete — the button reverts to ⌕ Detect text (enabled) when done. This may take a few seconds depending on image complexity.

  4. Review the proposals — green bounding-box overlays appear on the image marking detected text regions. The redaction regions counter in the panel updates to show how many regions were detected.

Expected Result

  • After scanning completes, at least one green bounding-box overlay is visible on the image (the counter increments).
  • The detected regions are proposals only — they are not applied until you click Apply & Export.
  • You can manually adjust, delete, or add additional redaction boxes before finalizing.
  • The OCR runs on the current viewport frame. For multi-frame series, scan each frame independently.
  • The detection works on monochrome medical images (CT, MR, CR, DX) with standard burned-in text patterns.