How to decide if a DICOM series is a 3D volume or a series of images?

Question

We are writing an importer for dicom files.

How is it generally fooled if a series of images forms 3D volumes or is only a series of 2D images?

Is there a universal way to solve this for most suppliers? I looked at the DICOM tags and could not find a visible solution.

Answer 1

The DICOM standard defines a UID for describing a hierarchy. This is top to bottom:

• Study UID - identifier of the study or scan session.
• The UID of the series is the same in the series obtained in one scan.
• Image UID - must be unique to any image.

A DICOM image saved by a standard compatible implementation should have all of these identifiers. If multiple images have the same SeriesUID, they are volumes (or time series) as defined in the standard. Some software, of course, is not up to standard, and you will have to look for other things, such as time stamps and patient position, but it is usually best to start by following the standard.

To streamline the series after its identification, GDCM (at the suggestion of malate) or dcmtkdicom are pretty well-established libraries.

Answer 2

I think you are looking for an algorithm to organize a DICOM dataset using Image Position (Patient) and Image Orientation (Patient).

A typical implementation can be found in GDCM

Please note that my answer may be completely unrelated to your specific DICOM instances, but since you did not indicate which SIP UID class you were dealing with, I just assumed that you are dealing with an old CT or MR image store

Answer 3

In MR you need to look for:

Type of acquisition MR (0018,0023). It has two enumerated values:

• 2D = frequency x phase
• 3D = frequency x phase x phase

I'm not sure about CT.

In most cases, the small answer is what you want to do (for example, arrange slices by position and orientation and process them in 3D mode through a multi-faceted reconstruction).

Answer 4

The patient position (0018, 5100) is a mandatory attribute of type 1 for both CT and MR modality. This attribute is VERY IMPORTANT for accurate interpretation of patient orientation.

Answer 5

Typically, a projection x-ray has a patient orientation attribute (0020, 0020), and the cross-sectional image should have the Image Position (0020, 0032) and Image orientation (0020, 0037) attributes, since they are required elements of type 1 of the Image Plane module (see section PS.7.6.2.1.1). However, the image of the localizer or scout included in the CT scan is not really a cross-sectional image, but a projected image and may contain attributes of the image position and image orientation. The same is true for MRI studies, where one or more sagittal or coronal images are usually recorded, of which axial images are prescribed. In this case, a different logic is needed to identify the localizer image. For example, a CT locator might use the string "LOCALIZER" for the value of the 3 attributes "Image Type".