Stability of Radiomics Features Using 4D-CT Across Radiomics Platforms and Contour Variations for Lung and Liver Tumors
X Wang1*, C Ma1, Y Li1,2, Y Zhang1, Y Zhang1, N Yue1, K Nie1, (1) Rutgers Cancer Institute of New Jersey, Rutgers Robert Wood Johnson Medical School, New Brunswick, NJ, (2) Saint Vincent Hospital, Worchester, MA
Presentations
PO-GePV-M-68 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: To evaluate the stability of radiomics features using 4D-CT as an alternative to test-retest CT-scans, and the agreement/consistency across radiomics platforms, different sites, and contouring variations.
Methods: 4D-CT images with 10 breathing phases were acquired for 10 patients with lung tumors and 10 patients with liver tumors. For each patient, two contours (GTV and GTV-1mm) were delineated on each breathing phase. GTV-1mm was generated by subtracting a 1mm inner margin from GTV to investigate the impact of contouring variation. 42 common radiomics features were extracted using three open-source radiomics platforms. The coefficient of variation (COV) of each feature was calculated from all phases to evaluate test-retest stability. The concordance correlation coefficient (CCC) and the intraclass correlation coefficient (ICC) of each feature were calculated to evaluate the agreement/consistency of the three radiomics platforms, lung and liver tumors, and different contours.
Results: The features are more stable in test-retest for liver tumors than lung tumors. 40 features out of 42 have a mean COV less than 0.5 for liver tumors on all three platforms, while only 26 features have a mean COV less than 0.5 for lung tumors on all three platforms. The three platforms concord (CCC>0.8) on 7 features for lung tumors, and 9 features for liver tumors, taking contouring variations into account. Regarding contouring variation, 21 features concord (CCC>0.8) on all three platforms for lung tumors, while 26 features concord on all three platforms for liver tumors.
Conclusion: The results show that the stability of radiomics features can be evaluated using 4D-CT as an alternative to test-retest CT-scans. The stability evaluation depends on the platforms, contouring accuracy and disease site. Overall, we are able to identify stable radiomics features in test-retest and across platforms and contouring variations, which may have potential to be used for radiomics-guided clinical studies.
Keywords
Feature Extraction, Feature Selection
Taxonomy
IM/TH- Image Analysis (Single Modality or Multi-Modality): Image processing
Contact Email