DICOM basics for volumetric data

4
votes

I have looked at some of the DICOM standard and the Wikipedia page (and all DICOM topics on SO) but without really digging into the file structure doc (ugh), I'm left confused what exactly is stored inside a DICOM file for a scan comprising 3D/4D/5D data. I only want an overview.

Let's take MRI as an example. Does the DICOM file contain

  1. a set of the raw 2D images taken from various angles
  2. a stack of slices forming a 3D voxel dataset
  3. a full 3D dataset

In other words, does DICOM include any post-processing on the raw images captured by the imaging machine?

As far as 4D, presumably it's simply a collection of multiple 3D datasets, each 'frame' is a separate dataset?

2
dicommedical
+1 I see the connection between point (1) in your question and your post on Physics SE - (1) would apply to CT scanning. - Antoni Parellada

2 Answers

5
votes

The DICOM "standard" has so many different options in it, along with vendor extensions, that most things are possible.

That said, most 3D DICOM datasets that I have encountered are 2D image stacks. For post-processing, it depends solely on the imaging machine.

If you're looking for a freely available DICOM library, you can try out GDCM, I've been fairly happy with it.

1
votes

The original DICOM MR format holds a single, 2D image. That format is still widely supported, quite likely what you have. If that is what you have, then you'll want to gather all of the images in a series to construct your volume.

A newer, enhanced MR image format is able to store an entire volume as a multi-frame image. If it does, that's your volume. I don't recall whether a single enhanced MR object has to be the entire series or not.

You can tell them apart by looking at the DICOM attribute SOP Class UID (0008,0016):

 Original MR:  1.2.840.10008.5.1.4.1.1.4   
 Enhanced MR:  1.2.840.10008.5.1.4.1.1.4.1

Above, when I said that you might need to gather "all of the images in a series" to build a volume, that means you are using the images that share the same value for the attribute Series Instance UID (0020,000E).

Notes:

  • DICOM purists will prefer the term "Image Information Object Definition" instead of the word "format" as I've been using it.
  • DICOM purists will object to saying that SOP Class UID defines the image format. For your purposes, it will work.