ESTRO 2020

Session Item

August 28
16:20 - 16:40
Jens Overgaard Legacy Award
Award lecture
23:00 - 23:00
Findable, Accessible, Interoperable, Reusable (FAIR) Quantitative Imaging Analysis Workflow


Findable, Accessible, Interoperable, Reusable (FAIR) Quantitative Imaging Analysis Workflow
Authors: Shi|, Zhenwei(1)*[];Fedorov|, Andrey (2);Hosny|, Ahmed (3);Parmar|, Chintan(3);Aerts|, Hugo(2,3);Wee|, Leonard(1);Dekker|, Andre(1);
(1)GROW School for Oncology and Developmental Biology- Maastricht University Medical Centre+, Department of Radiation Oncology MAASTRO CLINIC, Maastricht, The Netherlands;(2)Brigham and Women’s Hospital, Department of Radiation Oncology, Boston, USA;(3)Dana Farber Cancer Institute, Department of Radiation Oncology, Boston, USA;
Show Affiliations
Purpose or Objective

The objective of this study was to demonstrate the proof of concept FAIR Quantitative Imaging Analysis Workflow (FAIR-QIAW) on the top of DICOM data, which can make quantitative imaging FAIRer by generating a standardized DICOM representation of the annotation results. 

Material and Methods

Four datasets were already publicly available on XNAT ( The MMD dataset includes PET-CT images, RTSTRUCT of 22 NSCLC patients with 10 delineations of gross tumour volume (GTV) were drawn by 5 doctors. The RIDER dataset includes DICOM CT images and RTSTRUCT of 32 NSCLC patients, who underwent two thoracic CT scans within 15 minutes with the same CT scanner and imaging protocol. The HN1 dataset includes DICOM CT images and RTSTRUCT of 136 consecutive patients with SCC of the head and neck. The LUNG1 dataset includes DICOM CT images and RTSTRUCT 422 consecutive patients treated with (chemo)radiation. To provide the proof of concept of FAIR quantitative imaging, we only used CT image and GTV as the volume of interest (VOI). The processing workflow is shown in Figure-1. First, a JSON file was created, which incorporates the meta-information of segmentation coded by the DICOM standard and SNOMED ontology. Second, the Plastimatch library was used to convert the DICOM images to image volume and generate the binary mask of GTV, which were saved locally in the NRRD format and path information were stored in a CSV table for further query.  DICOM segmentation object (SEG) series containing the segmentation results were created using dcmqi library ( Third, three types of calculation were implemented using the data generated above: (1) a 2-year survival DNNs (deep-prognosis, packaged in a Docker container can calculate deep learning-based radiomics. (2) PyRadiomics can calculate radiomic features using a flexible configuration file. (3) PyRadiomics-dcm can generated DICOM structured reporting (SR) using the DICOM SEG files, DICOM images, Radiomics Ontology and related parameters. 


Totally 632 patients were included in this study. The FAIR-QIAW was developed and implemented using an AWS instance with Ubuntu 18.04, 4 virtual CPU, 16 GB memory and 200 GB volume space. A bash script was generated to install all relevant dependencies and packages and download DICOM data of 4 datasets from the XNAT repository automatically. The results comprised VOI binary masks, volume image, DICOM SEG of all VOIs, DNN results, radiomic features, and DICOM SR coded by ontologies. All the processes were implemented atomically.

Figure 1:. The diagram of FAIR Quantitative Imaging Analysis Workflow. Level-1 represents original DICOM data with metadata of ROI segmentation; Level-2 represents intermediate data generated by related tools, configuration parameters and domain ontology; Level-3 represents Docker computational components. Level-4 represents computational results generated by FAIR-QIAW. Level-5 represents learning application components.


We proposed analysis workflow to make the publicly available datasets FAIRer for quantitative imaging research, by generating DICOM SEG, reporting DICOM SR as well as providing conversion into other common standards. Furthermore, our program supports universal radiomics extraction using PyRadiomics and DNNs implementation via Docker container. Finally, the source codes allow other researchers to reuse the data.