Glossary

Quick overview of all terms used throughout MIDOM

Protocol

Deidentification Protocol. A full, abstract description of a deidentification procedure. Describes what to do with each DICOM Element, voxel data, private elements. Also describes which datasets to reject completely. It consists of four elements Tags, Filters, Pixel, Private. See the objects page as well.

Deidentifier

The concrete implementation of a Protocol. Takes a DICOM Dataset and then either aplies a transformation to it, or rejects it. Different deidentifiers can implement the same protocol.

DICOM Tag

The name for a single type of data in a DICOM Dataset. Like Modality, PatientAge, etc. All 4000+ official dicom tags are listed here. There are also private tags which can be created by any DICOM producer.

Each dicom tag has an associated Value Representation (VR).

Value Representation (VR)

The ‘data type’ of a DICOM element’s value. For example an int, a string, a date. Also contains more exotic datatypes like ‘Unlimited Text’. All DICOM VR’s are listed in the DICOM standard

DICOM Element

A DICOM Tag - Value pair. Like PatientName: A_Smith

DICOM Dataset

A set of DICOM elements. For extended description, see the objects page

ImageData element

A DICOM Element that contains image data. In most DICOM files this element is several times bigger than all other elements combined. The most common image data element is PixelData
(7FE0,0010) PixelData
But there are several others:
(7FE0,0008) FloatPixelData
(7FE0,0009) DoubleFloatPixelData
(0070,0022) GraphicData
(5400,0110) WaveformData
This list might not be complete. The criteria for what constitutes ‘image data’ are not completely set.

ImageData elements can contain two types of PHI: Burnt-in image PHI and Dynamic image PHI

Burnt-in image PHI

Burnt-in / Static image PHI is always in the same place in an image. Many DICOM-producing modalities, especially in Ultrasound, write PHI like patient name and date of birth into the image. For a specific vendor, model and dataset type, this information can always be found at the same X-Y coordinates.

Dynamic image PHI

Dynamic image PHI has no pre-determined place. It is not added to the image on purpose. Faces and implant serial numbers fall into this category.

Action

An intended change to a single DICOM Element. The change is expressed as an Action code.

Delta

An observed change to a single DICOM Element. The change is expressed as a delta code.

Delta set

A set of Deltas for a set of distinct DICOM elements. See the objects page.

PHI

Personal Health Information. Also called PII (Personally Identifiable information) or `PI` (Personal Information). MIDOM prefers the term PHI as this is more specifically defined in the health domain.

For medical imaging data, personal health information is found in many of the standard DICOM elements. Obvious ones are patient name and date of birth. Next to that there are many more. The DICOM standard lists more than 400 elements for which some kind of processing is required to deidentify.

In addition to elements, PHI can also be present in image data directly as Burnt-in image PHI or Dynamic image PHI

Pixel data

A special DICOM Element that contains the bytes for the image component of a DICOM dataset. This element often take up many times more data than all other elements combined. Its processing is done by the specialized pixel module

Private tag

DICOM private tags are custom data elements that aren’t part of the standard specification, allowing healthcare organizations to store proprietary or specialized information. Private tags enhance flexibility, but are a well known PII leak risk. They are handled by the specialized private module.

More information on private tag structure can be found in the DICOM standard (very tough read), or in the pydicom docs (more understandable).