pyradiomics labs

pyradiomics labs is a collection of exploratory/experimental features that are part of the repository, but are not part of the core functionality. We welcome user feedback about those features. Those scripts and features may change in the future.



This is an experimental script to support the use of pyradiomics with DICOM data.

The script will accept as input a directory with a single DICOM image study for the input image, and the file name pointing to a DICOM Segmentation Image (DICOM SEG) object.

The script will transparently convert the DICOM image into a representation suitable by pyradiomics using either plastimatch or dcm2niix.


  • medical image data usually comes in DICOM, and pyradiomics users often ask for help working with DICOM data
  • there are public collections of data on TCIA where segmentations are stored as DICOM SEG
  • the use of DICOM representation for radiomics features
    • introduces standardized formalism for the attributes that should be stored to accompany the features
    • allows to link results of calculations with the various ontologies describing the anatomy of the regions analyzed, and the features itself (e.g., the SR document produced by the script will utilize IBSI nomenclature to describe those features implemented in pyradiomics that have correspondence in IBSI)
    • allows to reference (by unique identifiers) the DICOM image series and DICOM segmentation used for feature calculation
    • enables harmonized representation of data for images, segmentations and features (i.e., same data management system can be used for all data types)
    • does not prevent the use of the results in software tools that are not DICOM-aware - dcmqi can be used to convert DICOM segmentations and DICOM SR with the measurements into non-DICOM representation (ITK-readable image formats for segmentations, and JSON for measurements); a separate tool is available to generate tab-delimited representation for DICOM attributes and measurements stored in those SRs:


  • plastimatch or dcm2niix for image volume reconstruction
  • dcmqi [1] [2] (build from fedb41 or later) for reading DICOM SEG and converting to a representation suitable by pyradiomics, and for storing the resulting features as a DICOM Structured Report, instantiating SR TID 1500
  • prior to using this script, you might want to sort your DICOM data such that individual series are stored in separate directories. You might find this tool useful for this purpose:
  • if you segmentations are not stored as DICOM SEG, you can use dcmqi for generating standard representation of those segmentations:


Example usage from command line:

$ python -h
usage: --input-image <dir> --input-seg <name> --output-sr <name>

Warning: This is a "pyradiomics labs" script, which means it is an experimental feature in development!
The intent of this helper script is to enable pyradiomics feature extraction directly from/to DICOM data.
The segmentation defining the region of interest must be defined as a DICOM Segmentation image.
Support for DICOM Radiotherapy Structure Sets for defining region of interest may be added in the future.

optional arguments:
  -h, --help            show this help message and exit
  --input-image-dir Input DICOM image directory
                    Directory with the input DICOM series. It is expected
                    that a single series is corresponding to a single
                    scalar volume.
  --input-seg-file Input DICOM SEG file
                    Input segmentation defined as aDICOM Segmentation
  --output-dir Directory to store the output file
                    Directory for saving the resulting DICOM file.
  --parameters pyradiomics extraction parameters
  --temp-dir Temporary directory
  --features-dict Dictionary mapping pyradiomics feature names to the IBSI defined features.
  --volume-reconstructor Choose the tool to be used for reconstructing image volume from the DICOM image series. Allowed options are plastimatch or dcm2niix (should be installed on the system). plastimatch will be used by default.

Sample invocation

$ python --input-image-dir CT --input-seg SEG/1.dcm \
   --output-dir OutputSR --temp-dir TempDir --parameters Pyradiomics_Params.yaml
dcmqi repository URL: revision: 3638930 tag: latest-4-g3638930
Row direction: 1 0 0
Col direction: 0 1 0
Z direction: 0 0 1
Total frames: 177
Total frames with unique IPP: 177
Total overlapping frames: 0
Origin: [-227.475, -194.775, -1223]
dcmqi repository URL: revision: 3638930 tag: latest-4-g3638930
Total measurement groups: 1
Adding to compositeContext: 1.dcm
Composite Context initialized
SR saved!

$ dsrdump OutputSR/
Enhanced SR Document

Patient             : interobs05 (#interobs05)
ENH: include pyradiomics identification and version
Study               : interobs05_20170910_CT
Series              : GTV segmentation - Reader AB - pyradiomics features (#1)
Manufacturer        : QIICR (, #0)
Completion Flag     : PARTIAL
Verification Flag   : UNVERIFIED
Content Date/Time   : 2018-10-09 15:38:01

<CONTAINER:(,,"Imaging Measurement Report")=SEPARATE>
  <has concept mod CODE:(,,"Language of Content Item and Descendants")=(eng,RFC5646,"English")>
  <has obs context CODE:(,,"Observer Type")=(121007,DCM,"Device")>
  <has obs context UIDREF:(,,"Device Observer UID")="">
  <has obs context TEXT:(,,"Device Observer Name")="pyradiomics">
  <has obs context TEXT:(,,"Device Observer Model Name")="2.1.0.post10.dev0+g51bc87f">
  <has concept mod CODE:(,,"Procedure reported")=(P0-0099A,SRT,"Imaging procedure")>
  <contains CONTAINER:(,,"Image Library")=SEPARATE>
    <contains CONTAINER:(,,"Image Library Group")=SEPARATE>
      <has acq context CODE:(,,"Modality")=(CT,DCM,"Computed Tomography")>
      <has acq context DATE:(,,"Study Date")="20170910">
      <has acq context UIDREF:(,,"Frame of Reference UID")="">


  <contains CONTAINER:(,,"Imaging Measurements")=SEPARATE>
    <contains CONTAINER:(,,"Measurement Group")=SEPARATE>
      <has obs context TEXT:(,,"Tracking Identifier")="Gross Target Volume">
      <has obs context UIDREF:(,,"Tracking Unique Identifier")=""
      <contains CODE:(,,"Finding")=(C112913,NCIt,"Gross Target Volume")>
      <contains IMAGE:(,,"Referenced Segment")=(SG image,,1)>
      <contains UIDREF:(,,"Source series for segmentation")="
      <has concept mod CODE:(,,"Finding Site")=(T-28000,SRT,"Lung")>
      <contains NUM:(,,"shape_MeshVolume")="7.255467E+04" (1,UCUM,"no units")>
      <contains NUM:(,,"Maximum 3D diameter")="7.491328E+01" (1,UCUM,"no units")>
      <contains NUM:(,,"shape_Maximum2DDiameterSlice")="6.767570E+01" (1,UCUM,"no units")>
      <contains NUM:(,,"Elongation")="7.993260E-01" (1,UCUM,"no units")>
      <contains NUM:(,,"shape_MinorAxisLength")="4.699969E+01" (1,UCUM,"no units")>
      <contains NUM:(,,"Flatness")="6.517569E-01" (1,UCUM,"no units")>
      <contains NUM:(,,"shape_Maximum2DDiameterColumn")="6.746851E+01" (1,UCUM,"no units")>
      <contains NUM:(,,"Surface to volume ratio")="1.572168E-01" (1,UCUM,"no units")>
      <contains NUM:(,,"shape_Maximum2DDiameterRow")="6.072891E+01" (1,UCUM,"no units")>
      <contains NUM:(,,"shape_VoxelVolume")="7.285600E+04" (1,UCUM,"no units")>
      <contains NUM:(,,"Sphericity")="7.375024E-01" (1,UCUM,"no units")>
      <contains NUM:(,,"Surface area")="1.140681E+04" (1,UCUM,"no units")>
      <contains NUM:(,,"shape_MajorAxisLength")="5.879915E+01" (1,UCUM,"no units")>
      <contains NUM:(,,"shape_LeastAxisLength")="3.832275E+01" (1,UCUM,"no units")>
      <contains NUM:(,,"Small zone emphasis")="7.384502E-01" (1,UCUM,"no units")>
      <contains NUM:(,,"glszm_SmallAreaLowGrayLevelEmphasis")="3.381883E-03" (1,UCUM,"no units")>
      <contains NUM:(,,"Normalised grey level non-uniformity")="3.136554E-02" (1,UCUM,"no units")>
      <contains NUM:(,,"glszm_SmallAreaHighGrayLevelEmphasis")="5.478214E+02" (1,UCUM,"no units")>
      <contains NUM:(,,"Large zone emphasis")="3.873234E+03" (1,UCUM,"no units")>



Please post your feedback and questions on the pyradiomics email list.


[1]Herz C, Fillion-Robin J-C, Onken M, Riesmeier J, Lasso A, Pinter C, Fichtinger G, Pieper S, Clunie D, Kikinis R, Fedorov A. dcmqi: An Open Source Library for Standardized Communication of Quantitative Image Analysis Results Using DICOM. Cancer Research. 2017;77(21):e87–e90
[2]Fedorov A, Clunie D, Ulrich E, Bauer C, Wahle A, Brown B, Onken M, Riesmeier J, Pieper S, Kikinis R, Buatti J, Beichel RR. (2016) DICOM for quantitative imaging biomarker development: a standards based approach to sharing clinical data and structured PET/CT analysis results in head and neck cancer research. PeerJ 4:e2057