Inability to access a particular set of DICOM tag through normal pydicom methods/attributes.

[2ManySmells]

Description of Bug

I have a CT modality DICOM file. In it are a collection of DICOM tags prefaced by 'Origin xxxxx' that cannot be accessed and throws a bunch of exceptions.

If i use dcm.dir() or dcm.keys() I can observe that this DICOM tag does exist. Attached is a picture that shows the existence of the said tag(s). The DICOM file was extracted through the dcmread method in pydicom. For this example I will be using the Tag (6000, 0050) Overlay Origin

If I use dcm.dir() or dcm.keys(), both methods verify the existence of the said tag.

However using the dcm[keyword] method will throw this error, shown here below:

Using the method, dcm[0x6000, 0x0050].keyword will return an empty string, specifically: ''

The only way to access the tag and it's value is by accessing it through dcm[element], so in this case dcm[0x6000, 0x0050]

Expected behaviour

The expectation if there exists a DICOM tag presented by either dir() or keys() method. The tag should be accessable by all other methods.

Reproducing the error

Currently on our project, as long as the CT file contains this Overlay xxxx tag we would experience this bug.

• read in file through pydicom.dcmread(fp, stop_before_pixels=True)
• check if the Tag exists through FileDataSet.dir()
• access tag through FileDataSet['OverlayOrigin'] (example)

Environment

Currently python 3.9.1
pydicom 2.1.2

[darcymason]

This is a case of an old-style DICOM way of doing things (now handled by Sequences) where certain tags can be repeated. If you look in pydicom's RepeatersDictionary you will see this defined as

'60xx0050': ('SS', '2', "Overlay Origin", '', 'OverlayOrigin'),

Notice the 'xx' - DICOM used to allow multiple tags to map to the same description, and therefore later with the advent of keywords, to the same keyword.

I don't recall the details in pydicom - we could dig into it - but in general pydicom would not be able to access these by keyword, because there could be many data elements with that same keyword, e.g. (6000,0050), (6002, 0050), (60aa, 0050), ... anything that matches the 'xx'. The tag number is the only way to guarantee access to the intended data element.

It's not something we've looked at in a long time - we should at least document this better, or perhaps the code could be modified to search for all matches and generate an error if more than one found, otherwise just return the one. But for setting data elements using the keyword, it would be harder - probably the user would somehow have to specify the 'xx'.

So, it is a bit of a grey area - but certainly the documentation should point to this. Even better, a detailed error message on attempting keyword access for items in the RepeatersDictionary should point out the tag number is necessary. The latter sounds like a good addition to the code.

[scaramallion]

Setting repeating group elements is covered in the basics tutorial, as is a summary of what they are.

Because of the lack of a unique keyword, this won’t work for private or repeating group elements. So for those elements stick to the Dataset[group number, element number] method.

[2ManySmells]

@darcymason Appreciate the clarification regarding items in the RepeatersDictionary, will keep that in mind if we meet a similar issue in the future.

@scaramallion My bad, I probably skimmed over the basic tutorial and didn't realise that pydicom does have a tutorial on handling Overlay Data (06xx, 3000). Thanks for the explanation!