DICOM De-Identification
DICOM anonymization (a type of de-identification) is the commonest workflow in ChRIS. This page describes the DICOM anonymization pipeline.
Background
The DICOM file format is the international standard for how medical images are produced, transmitted, and stored. Hence, de-identification of DICOM data is the first step of any imaging research analysis.
Protected health information (PHI) can be found in DICOM files in three forms:
- Metadata as DICOM tags, e.g. PatientName, PatientSex, PatientBirthDate, etc.
- Burned text, where PHI appears as text in the image (the PixelData).
- Identifiable anatomy e.g. the face which is present on head MRI scans.
Currently, metadata anonymization and burned text removal are available as ChRIS pipelines.
DICOM Tag Anonymization
PHI found in DICOM tags can be anonymized by deleting those tags. This can be done:
- Manually
- Automatically, using an "allow-list" (i.e. remove all tags except for a few)
- Automatically, using a "deny-list" (i.e. keep all tags except for a few)
Generally, "allow-list" approaches are considered safer. The risk of accidental PHI leakage is lower using an "allow-list" policy because it is stricter.
DICOM De-Identification in ChRIS
The following ChRIS plugins are available:
| Plugin Name | Description |
|---|---|
| pl-dicomize | Among its multiple purposes, does DICOM anonymization by removing all DICOM tags except for the image data. |
| pl-dicom_headerEdit | Edits specified DICOM tags. This is a "deny-list" approach. |
| pl-dcm_textlocr | Detects burned text PHI using optical-character recognition (OCR). |
| pl-image_textRemove | Detects and removes burned-in text from images using OCR. |
These plugins are the building blocks of our DICOM de-identification pipelines. We have many pipelines tailored for specific workflows (e.g. ultrasound anonymization, anon → analyze → upload, ...), please contact us for more information.