Figure 1.
Schematic diagram depicting the overview of the analysis.
A: First, we performed five manual delineations and six 3D-Slicer segmentations (three observers twice) on twenty lung tumors. B: Second, fifty-six radiomic features quantifying tumor intensity, texture and shape were extracted from these segmentations. C: Third, the resulting feature matrices were compared for robustness of the feature values.
Figure 2.
Feature wise comparison of Intra-class correlation coefficients (ICC) between manual and 3D-Slicer segmentations.
A: First order statistics features. B: Shape based features. C: Textural features.
Figure 3.
Box-plot comparing intra- and inter-observer reproducibility (ICC) of radiomic features.
High inter- and intra- observer reproducibility (ICC) was observed for 3D-Slicer segmentations compared to the inter-observer reproducibility (ICC) of manual delineations. From left the first box refers to the manual inter-observer reproducibility (ICC), second and third boxes refer to the inter-observer reproducibility (ICC) of two different 3D-Slicer segmentation runs. Remaining three boxes refer to the intra-observer reproducibility (ICC) of 3D-Slicer segmentations.
Figure 4.
Comparison of normalized feature range between manual and 3D-Slicer segmentation groups.
Radiomic features derived from 3D-Slicer segmentations had significantly smaller and overlapping range compared to that from manual delineations.