...
This dataset was used by the NCI's Quantitative Imaging Network (QIN) PET-CT Subgroup for their project titled: Multi-center Comparison of Radiomic Features from Different Software Packages on Digital Reference Objects and Patient Datasets. The purpose of this project was to assess the agreement among radiomic features when computed by several groups by using different software packages under very tightly controlled conditions, which included common image data sets and standardized feature definitions. The image datasets (and Volumes of Interest – VOIs) provided here are the same ones used in that project and reported in the publication listed below. In addition, we have provided the individual feature value results for each image dataset and each software package that was used to create the summary tables (Tables 2, 3 and 4) in that publication. For that project, nine common quantitative imaging features were selected for comparison including features that describe morphology, intensity, shape, and texture and that are described in detail in the International Biomarker Standardisation Standardization Initiative (IBSI, https://arxiv.org/abs/1612.07003).
There are three datasets provided – two image datasets and one dataset consisting of three excel spreadsheets containing feature values.
- The first image dataset is a set of three Digital Reference Objects (DROs) used in the project, which are: (a) a sphere with uniform intensity, (b) a sphere with intensity variation (c) a nonspherical (but mathematically defined) object with uniform intensity. These DROs were created by the team at Stanford University and are described in (Jaggi A, Mattonen SA, McNitt-Gray M,Napel S. Stanford DRO Toolkit: digital reference objects for standardization of radiomic AQ: F features. Tomography. 2019;6:–.) and are a subset of the DROs described in (link to DRO wiki page). Each DRO is represented in both DICOM and NIfTI format and the VOI was provided in each format as well (DICOM Segmentation Object (DSO) as well as NIfTI segmentation boundary).
...
- The second image dataset is a set of 10 patient CT scans which is a subset of the QIN multi-site collection of Lung CT data with Nodule Segmentations (http://doi.org/10.7937/K9/TCIA.2015.1BUVFJR7) created previously. Specifically, the same 10
...
- cases selected from the LIDC-IDRI dataset that were used in that previous study were used in this study. As in that previous study, a single lesion from each case was identified for analysis. That previous study generated VOIs using algorithms
...
- from three academic institutions and each method performed three repeat runs on each nodule. For this study, and to eliminate one source of variability in our project, one of the VOIs previously created for each lesion was identified and all sites used that same
...
- VOI definition. The specific VOI chosen for each lesion was the first run of the first algorithm (algorithm 1, run 1). As in that prior project, both DICOM and NIfTI formats were created for each image dataset and the VOI was provided in each format as
...
- well (DICOM Segmentation Object (DSO) as well as NIfTI segmentation boundary).
...
...
- The third dataset is a collection of three excel spreadsheets, each of which contains the raw feature values and the summary tables reported in the publication below. These tables are:
...
...
a.
...
DRO results: This contains the original feature values obtained for each software package for each DRO as well as the table summarizing results across software packages (Table 2 in the publication).
...
...
b.
...
Patient Dataset results: This contains the original feature values for each software package for each patient dataset (1 lesion per case) as well as the table summarizing results across software packages and patient datasets (Table 3 in
...
the publication).
...
...
c.
...
Harmonized GLCM Entropy Results: This contains the values for the “Harmonized” GLCM Entropy feature for each patient dataset and each software package as well as the summary across software packages (Table 4 in the publication).
Acknowledgements
The authors gratefully acknowledge the following sources of support:
...