DO-NOT-USE dicom multiframe to single frame conversion

README.md

this gist is used as a supporting doc, in response to 
https://github.com/pydicom/pydicom/discussions/1705

do-not-use.py

import os
import sys
import pydicom
essential_group_list = [0x0008,0x0010,0x0018,0x0020,0x0028]

def main(input_file,output_folder):
    
    os.makedirs(output_folder,exist_ok=True)
    
    ds = pydicom.dcmread(input_file)
    number_of_frames = int(ds.NumberOfFrames)
    
    # requirements
    assert(number_of_frames > 1)
    assert(ds.pixel_array.shape[0]==number_of_frames)
    assert('SharedFunctionalGroupsSequence' in ds)
    assert('PerFrameFunctionalGroupsSequence' in ds)
    assert(len(ds.SharedFunctionalGroupsSequence)==1)
    assert(len(ds.PerFrameFunctionalGroupsSequence)==number_of_frames)

    img = ds.pixel_array.squeeze()
    
    for idx in range(number_of_frames):
        target_file = os.path.join(output_folder,f'foobar.{idx:05d}.dcm')

        file_meta = pydicom.dataset.FileMetaDataset()
        file_meta.MediaStorageSOPClassUID = pydicom.uid.SecondaryCaptureImageStorage
        file_meta.MediaStorageSOPInstanceUID = pydicom.uid.generate_uid()
        file_meta.TransferSyntaxUID = pydicom.uid.ImplicitVRLittleEndian # pydicom.uid.ExplicitVRLittleEndian

        tmp_ds = pydicom.dataset.Dataset()
        tmp_ds.file_meta = file_meta
        tmp_ds.is_little_endian = tmp_ds.file_meta.TransferSyntaxUID.is_little_endian
        tmp_ds.is_implicit_VR = tmp_ds.file_meta.TransferSyntaxUID.is_implicit_VR
        
        essential_group_list = [0x0020,0x0028] # barebones...
        for group in essential_group_list:
            for k,v in ds.group_dataset(group).items():
                tmp_ds[k]=v
        
        foo = ds.SharedFunctionalGroupsSequence[0]
        bar = ds.PerFrameFunctionalGroupsSequence[idx]

        for tag_name in ['RescaleIntercept','RescaleSlope','RescaleType']:
            if tag_name in foo[(0x0028,0x9145)][0]:
                value = foo[(0x0028,0x9145)][0][tag_name].value
                setattr(tmp_ds,tag_name,value)

        del tmp_ds.NumberOfFrames
        tmp_ds.InstanceNumber = idx+1
        tmp_ds.PixelData = img[idx,:].squeeze().tobytes()
        tmp_ds.save_as(target_file,write_like_original=False)

if __name__ == "__main__":
    input_file = sys.argv[1]
    output_folder = sys.argv[2]
    
    raise NotImplementedError()
    
    main(input_file,output_folder)

'''
** DISCLAIMER **
i definitely would not use the above. 
only tested using 1 hologic tomosythesis multiframe file.

Clients should acknowledge and accept they are okay having private tags nuked.
Further, to prove this method works,
you'll need to gather dicom file flavors from your clients/institutions and then implement and verify 
the output dicom files can be opened by downstream workstations

'''

NOTES.md

So how do you split multi frame to single frame dicom images?

First you should google the current top results in google
keywords I used: "split dicom multiframe to single frame"
If you don't like the results you can check the below related links:


i reserve all rights to not respond to questions here XDDDD
please go to stackoverflow and ask your questions there!

--

• relevant discussion in pydicom ! pydicom/pydicom#1705

• long thread: https://groups.google.com/g/comp.protocols.dicom/c/h-8rj9ObBMM?pli=1

• wat? https://patents.google.com/patent/US8041156B2/en

• existing tools:

  • dclunie's dicom3tools dcuncat (you will need to compile it yourself)

  https://github.com/QIICR/dicom3tools/blob/master/appsrc/dcfile/dcuncat.cc https://github.com/QIICR/dicom3tools/blob/master/INSTALL original site - https://www.dclunie.com/dicom3tools.html

  • dcm4chee's emf2sf (java yucky, haven't tried)

  https://github.com/dcm4che/dcm4che/blob/master/dcm4che-tool/dcm4che-tool-emf2sf/README.md

  • gdcm's gdcmtar (tried, did not work for 1 test scan - hologic tomosythesis)

  https://gdcm.sourceforge.net/html/gdcmtar.html

  • work-in-progress over at highdicom

  https://github.com/dcm4che/dcm4che/blob/master/dcm4che-tool/dcm4che-tool-emf2sf/README.md

  • last resort: copy and start hacking using my below do-not-use.py

personal relevant gist
https://gist.github.com/pangyuteng/2fbaea6b1eebb310fe46eef4d8f7e18f