DICOM Enhanced Multi frame for data exchange - QIICR/ProjectIssuesAndWiki GitHub Wiki

Objective

Explore the use of DICOM enhanced storage objects for exchange of multi-dimensional datasets used by QIN projects.

Support the following workflow: multi-slice DICOM (3d or 4d) -conversion-tool-> multiframe DICOM -read-> Slicer -write-> multiframe DICOM

Specific tasks

1. Evaluate support provided for DICOM enhanced storage by various libraries (GDCM, dicom3tools, DCMTK, Slicer, ITK, Matlab)
2. Identify sample datasets
3. Create enhanced storage objects from relevant QIN datasets (3d and 4d datasets, BWH DBP)
4. Test and debug ITK IO pipeline support of enhanced storage DICOM objects (MRI only for now)

Status

Available tools for single-frame to multi-frame conversion and v.v.

• MultiFrameImageFactory from pixelmed
  • Usage example: java -cp pixelmed.jar -Xms512m -Xmx512m -Djava.awt.headless=true com.pixelmed.dicom.MultiFrameImageFactory ${inputPath} ${outputPath}
  • SetOfFrameSets
    • Usage example: java -cp pixelmed.jar -Xms512m -Xmx512m -Djava.awt.headless=true com.pixelmed.dicom.SetOfFrameSets ${inputPaths}
• dcmulti from dicom3tools software (code: [dcmulti.cc] (https://github.com/mehrtash/dcmdump/blob/master/dcmulti.cc)) WARNING: the use of this tool is strongly discouraged by its author David Clunie. Quoting from private communication with David:

I have an ancient "dcmulti" tool that does a similar job, but it is pretty horrible, and I extensively hacked it to produce the NEMA multiframe test CT and MR objects many years ago ... I have not updated it to clean up its output or to add specific support for the new Sup 157 SOP Classes. It might be a source of interest in terms of how NOT to write a tool to do this :)

I see from the Wiki page that someone has already made the mistake of trying to use my horrible dcmulti tool :)

• Usage example: dcmulti path/to/files/[0-n]* -v -sortby ImagePositionPatient > output.dcm
• gdcmtar from GDCM library (code: gdcmtar.cxx)
• emf2sf from dcm4che (code: Emf2sf.java )

Multi-frame Instances for BWH QIN sample data

Conversion using dcmulti

• Replicate ImageOrientationPatient to be consistent for all frames (deviations on the order of numerical precision will result in frames not treated as corresponding to a single multi-frame object) (helper script)

• conversion of prostate DCE sequence:

  dcmulti -derivedurationfromtiming -makestack -temporalposition
  -dimension StackID FrameContentSequence
  -dimension InStackPositionNumber FrameContentSequence
  -dimension TemporalPositionIndex FrameContentSequence
  -sortby ImagePositionPatient
  -of /tmp/PkModeling_verification_Case1_DCE_series_mf_
  dcmulti.dcm PkModeling_verification/Case1/DCE_series/*.dcm

• conversion of prostate T2w sequence:

  dcmulti -sortby -makestack ImagePositionPatient
  -dimension StackID FrameContentSequence
  -dimension InStackPositionNumber FrameContentSequence
  -of t2_dcmulti.dcm 5-T2_AX/*dcm

Issues:

1. UIDs are not initialized, since no valid root UID is available - fixed by using 2.25.* (fixup tool below). Is this a proper way to address it?
2. SOP Class is EnhancedMR, not Legacy converted - Legacy converted is not recognized by David's tools - added IOD to Pixelmed (TODO look in more detail)
3. Additional issues identified during validation by Pixelmed/DicomImageViewer and dciavdfy - need to fix / check if those can be taken care of by using legacy converted
4. DCMTK reader fails to read multiframe, since pixel spacing is encoded in per-frame functional group, and the DCMTK IO does not handle this case. GDCM can read, but only if the frames are ordered by ImageOrientationPatient. Z spacing is derived from IPP, but if frames are not in IPP order, no re-ordering is done (source code). 4d multiframe is not read by either DCMTK or GDCM readers.
5. Both GDCM and DCMTK fail to read LegacyConvertedEnhancedMRImageStorage SOP class

• Post-conversion fixups and ITK IO tests: https://github.com/fedorov/MultiFrameTests
• Pixelmed legacy converted enhanced MR storage IOD added here

Converted BWH-QIN sample datasets: http://slicer.kitware.com/midas3/folder/1999 (no errors on validation with pixelmed validator).

Sample Datasets

Multiframe

• NEMA DICOM Multiframe sample data (features spreadsheet)
• Vendor-supplied samples
  • Philips
  • Siemens
  • GE
  • GE BreastTomo
• Toshiba CT brain perfusion (not 100% DICOM compliant, but pretty close, has each time slice in separate instance/file i.e., set of 3D objects rather than 1 4D object)
• Vegetables - Philips

General Purpose

1. GDCM wiki sample dataset page
2. OsiriX DICOM image sets
3. XNAT Central public projects

References

1. Supplement 49: Enhanced MR Image Storage SOP Class
2. Supplement 58: Enhanced CT Image Storage SOP Class
3. [Clunie D., Erickson B.J. The New Enhanced Multiframe CT and MR DICOM Objects] (http://www.dclunie.com/papers/SS3DICOMObjectsSyllabus.pdf)
4. Supplement 157: Multi-frame Converted Legacy Images
5. Dec 16, 2013 hangout discussion minutes
6. IHE Diffusion Profile
7. IHE Perfusion Profile
8. Proposed changes to IHE Perfusion Profile (affects organization 3D/4D, currently out for ballot, will be adopted)