DICOM De-identification Profile

The foundational profile defined in DICOM PS3.15 Annex E. It specifies a list of attributes that must be blanked or replaced with pseudonyms to achieve a baseline level of anonymization. Key actions include:

• Removing all Patient Identification tags (e.g., Name, ID, Birth Date)
• Removing Institutional Identity tags (e.g., Institution Name, Address)
• Removing Equipment Identity tags (e.g., Device Serial Number)
• Retaining Patient Characteristics (e.g., Age, Sex, Weight) which are critical for research This profile is the starting point for most clinical trial and research data pipelines.

The fundamental addressable unit of data within a DICOM object. De-identification profiles operate by specifying actions on specific tags.

• Structure: A 32-bit identifier composed of a Group Number and an Element Number (e.g., (0010,0010) for Patient Name).
• Action: The DICOM Part 15 profile defines whether a tag must be removed, replaced, or retained.
• Private Tags: Odd-numbered groups (e.g., 0009,xx01) are vendor-specific and often contain hidden PHI, requiring aggressive scrubbing.

The irreversible process of removing all identifying information from a DICOM data set to prevent re-identification. This is the primary goal of the de-identification profile.

• Header Scrubbing: Modifying or deleting metadata tags like Patient Name (0010,0010) and Institution Name (0008,0080).
• Pixel Data: Requires detecting and redacting Burned-in Annotations—text or graphics permanently rendered into the image pixels.
• Longitudinal Consistency: A challenge where a single patient's multiple studies must retain a consistent, anonymous identifier for research.

A reversible alternative to anonymization where identifying data elements are replaced with artificial identifiers or pseudonyms. This allows longitudinal data linkage while protecting the original identity.

• Master Patient Index (MPI): Requires a secure, separate database to map the original Patient ID to the generated pseudonym.
• Consistency: The same pseudonym must be generated for the same patient across multiple visits and modalities.
• Use Case: Essential for clinical trials where tracking a patient's response over time is critical, but direct identifiers must be hidden from researchers.

Patient-identifying text or graphics that are permanently rendered into the pixel data of an image, not stored as separate metadata tags. This is a critical failure point for de-identification.

• Origin: Often added by older modalities or ultrasound machines that burn patient demographics directly onto the image frame.
• Detection: Requires Optical Character Recognition (OCR) or deep learning object detection models to locate regions of interest.
• Redaction: The identified region must be overwritten with a black box or blurred, ensuring the underlying pixel data is destroyed.

A specific de-identification standard defined by the HIPAA Privacy Rule (not DICOM itself, but a parallel regulatory framework). It lists 18 specific identifiers that must be removed.

• Identifiers: Includes names, dates (except year), telephone numbers, and full-face photographic images.
• DICOM Mapping: The DICOM Part 15 profile is designed to satisfy the Safe Harbor method by mapping HIPAA identifiers to specific DICOM tags.
• Expert Determination: The alternative HIPAA method, requiring a qualified statistician to certify that the risk of re-identification is very small.

A mandatory technical document published by a medical device vendor. For de-identification, it specifies exactly which profiles and options the system supports.

• Profile Declaration: The statement must declare if it supports the Basic Application Level Confidentiality Profile or more advanced options.
• Attribute List: It provides a detailed table of all DICOM tags the system modifies, retains, or removes during its de-identification process.
• Integration Check: This document is the primary resource for an integration engineer to verify that a tool's de-identification logic matches the site's specific privacy requirements.