Skip to content

8.3 Blur faces

On some head CT or MR exams, facial features could allow recognition even after tags are cleaned. Face de-identify blurs the face region and lets you review quality before keeping the result.

Demo series

CT_Head_With_Contrast — same series as 8.2 Harmonize names
(tests/controller/assets/test_dcm_files/CT_Head_With_Contrast).

Prefer Harmonize first so anatomy hints improve eligibility, then open face blur on this series.

Goal

On CT_Head_With_Contrast, run face blur in Gaussian mode, review QA, and save if appropriate.

Before you start

  1. Complete AI Features setup — face models and academic license.
  2. Import CT_Head_With_Contrast (and ideally finish Harmonize in 8.2).
  3. Open the series in Series View.

Workflow on CT_Head_With_Contrast (Gaussian)

  1. Open CT_Head_With_Contrast in Series View.
  2. Start Face blur / preview.
  3. Set blur mode to Gaussian (common default for this demo).
  4. Review side-by-side: current image vs proposed blur; green outline shows the face region.
  5. Check QA PASS vs QA FAIL (pixels changed outside the mask).
  6. Save to keep, or discard.

Face Blur Gaussian on CT_Head_With_Contrast

Other blur modes

Batch and Series View may also offer median, pixelate, or fill noise. The documented demo uses Gaussian only.

Batch

See 8.4 Run on many studies. Non-head series are skipped.

What good looks like

  • Face covered on CT_Head_With_Contrast; anatomy outside the mask unchanged.
  • Dataset Face blur status updates.

If it fails

  • Not a head series / insufficient face mask → skip or run Harmonize first (8.2).
  • Already applied → will not re-blur unless you use a deliberate re-run path.
  • QA FAIL → do not save; adjust mode or inspect mask.
  • Tools greyed out → AI Features.

Next steps

Continue with 8.4 Run on many studies to batch these tools across a cohort.