Effects of Interobserver Variability on 2D and 3D CT- and MRI-based Texture Feature Reproducibility of Cartilaginous Bone Tumors

Purpose or Learning Objective: To investigate the influence of interobserver manual segmentation variability on the reproducibility of bidimensional (2D) and volumetric (3D) unenhanced computed tomography (CT)- and magnetic resonance imaging (MRI)-based texture analysis.

Methods or Background: This retrospective study included 30 patients with cartilaginous bone tumors (10 enchondromas, 10 atypical cartilaginous tumors, and 10 intermediate- to high-grade conventional chondrosarcomas). Three radiologists independently performed manual contour-focused segmentation on unenhanced CT and T1-weighted and T2-weighted MRI by drawing both a 2D region of interest (ROI) on the slice showing the largest tumor area and a 3D ROI including the whole tumor volume. Additionally, a marginal erosion was applied to both 2D and 3D segmentations to evaluate the influence of segmentation margins on feature reproducibility. A total of 783 and 1,132 features were extracted from the original and filtered images and volumes, respectively. The intraclass correlation coefficient ≥ 0.75 indicated good to excellent interobserver reliability and defined feature stability.

Results or Findings: In 2D contour-focused versus margin shrinkage segmentation, the rates of stable features were 74.7% (585) versus 71.7% (561), 77.1% (604) versus 76.1% (596), and 95.7% (749) versus 96.4% (755) for CT and MRI T1-weighted, and T2-weighted images, respectively (p = 0.343). In 3D contour-focused versus margin shrinkage segmentation, they were 86.6% (980) versus 83.7% (947), 80.0% (906) versus 71.5% (809), and 95.0% (1,075) versus 65.7% (744) for CT and MRI T1-weighted, and T2-weighted volumes, respectively (p < 0.001). In 2D versus 3D segmentation, matching stable features derived from CT and MRI were 65.8% (515) versus 68.7% (778), and those derived from T1-weighted and T2-weighted images were 76.0% (595) versus 78.2% (885), respectively (p = 0.191 and 0.285).

Conclusion: Radiomic features of bone cartilaginous tumors extracted from 2D and 3D segmentations on CT and MRI examinations are reproducible, although some degree of interobserver segmentation variability exists and highlights the need for reliability analysis in radiomic studies. Margin shrinkage did not improve feature robustness compared with contour-focused segmentation.

No conflict of interest has been declared by the author(s).

Publication History

Article published online:
03 June 2021