Home / DICOM redaction software
DICOM Redaction Software
Patient identifiers in a DICOM study live in two places at once: burned into the pixels, and written into the header tags. Redactor handles both, as separate operations, across multi-frame studies — and the output stays DICOM.
Why It Is Different
2 surfaces
Pixels and tags, de-identified separately
Every frame
Multi-frame studies processed one frame at a time
Air-gapped
Studies need never leave your network
What Makes DICOM Hard
Identifiers live in two places
The modality burns patient text into the pixel data, often in a corner of every frame. The same identifiers are also written into the DICOM header as textual tags. Clean one surface and the other still identifies the patient. Both are handled, as deliberately separate operations.
Not just the first frame
A multi-frame study can carry identifiers anywhere in the series, so anything that samples one image and moves on passes a study that is still identifiable. Every frame is processed.
The study has to stay a study
Flattening to JPEG or PNG strips the series and study structure a PACS needs. The output remains DICOM, with that structure preserved.
Clinical text is different
Identifiers in a medical record take forms they do not take in a contract or a police report. Clinical text is de-identified with a model trained on the i2b2 clinical corpus rather than a general-purpose PII recogniser.
Imaging cannot leave
Studies are among the most tightly held data any organisation has.
Metadata beyond the DICOM tags
— file attributes carried alongside the study — is stripped or rewritten on export.
What DICOM Redaction Covers
Burned-in patient text
Identifiers written into the pixels by the modality
Header tag identifiers
Patient name, ID, birth date and address in the DICOM tags
Institution and physician
Referring and performing provider fields
Clinical narrative
PHI in the accompanying report, via a clinical model
Multi-frame series
Every frame of the study, not only the first
Accession and study IDs
Study, series and accession references
File metadata
Attributes carried alongside the study, stripped on export
Masking styles
Blur, pixelate or solid fill
Output remains DICOM, with study and series structure preserved. Available on shared SaaS, dedicated, private cloud, on-premises and air-gapped.
How It Works
Ingest
.dcm, .dic and .dicom are recognised as a distinct medical-imaging content type, not converted to a general image format.
Process frame by frame
Each frame of a multi-frame study is treated as one page, so burned-in text and objects are detected across the whole study.
De-identify the header separately
Extraction walks the DICOM tag set and returns the textual, PII-relevant tags, skipping elements that carry pixel data so imaging content is handled by image redaction instead.
Redact the pixels
Pixel-burned identifiers are obscured frame by frame, with blur, pixelate or solid fill.
Compared With Flattening to Images
| Flatten to JPEG or PNG | Redactor |
|---|---|
| Study and series structure lost; no longer a DICOM study | Structure preserved, output remains DICOM |
| Header tags carried over or discarded wholesale | Textual tags de-identified as a separate, deliberate operation |
| First frame sampled, later frames unchecked | Every frame of a multi-frame study processed |
| Clinical narrative run through a general PII model | Run through a model trained on clinical text |
| Studies uploaded to a cloud service to be processed | Runs on-premises or air-gapped, inside your network |
Where It Runs
For imaging, detection has to find identifiers in two places at once, and the content usually cannot leave the building while it does. In an air-gapped deployment the entire AI stack is local — transcription, vision, OCR, PII and language models — with no outbound inference call.
FAQ
DICOM Redaction Software questions, answered
Why is DICOM different from redacting a normal image?
Patient identifiers live in two places at once: burned into the pixel data by the modality, and written into the DICOM header as textual tags. Cleaning one leaves the other, and a tool that flattens the study to JPEG destroys the series structure a PACS needs.
Does it check every frame of a multi-frame study?
Yes. Each frame is processed in turn, because burned-in text does not necessarily appear on the first image. Anything that samples one frame and moves on will pass a study that is still identifiable.
Is the output still a DICOM file?
Yes. Study and series structure are preserved, so the de-identified study remains usable in the systems that expect DICOM.
How is the clinical report handled?
Clinical narrative is de-identified with a model trained on the i2b2 clinical corpus rather than a general-purpose PII recogniser, because identifiers in a medical record take forms they do not take in other documents.
Can studies be de-identified without leaving our network?
Yes. Redactor runs on-premises and fully air-gapped, with local models for transcription, vision, OCR and PII detection, so no content and no inference request leaves the deployment.
Which masking styles apply to DICOM?
Blur, pixelate or a solid fill, as elsewhere.
Redact a Study Without It Leaving Your Network
Send us a de-identified sample, or run it yourself on-premises. We will show you both surfaces: the pixels and the tags.