Table 2.
Number of radiomic features with excellent, moderate and poor robustness by feature class in multi-slice analysis.
| Feature class | Excellent robustness ICC ≥ 0.9 |
Moderate robustness 0.5 ≤ ICC < 0.9 |
Poor robustness ICC < 0.5 |
|---|---|---|---|
| Multi-slice Analysis: Original, fixed bin width = 25, B-spline interpolation | |||
| First order (n = 18) | 10 (55.6%) | 6 (33.3%) | 2 (11.1%) |
| GLCM (n = 24) | 17 (70.8%) | 7 (29.2%) | 0 (0.0%) |
| GLDM (n = 14) | 5 (35.7%) | 7 (50.0%) | 2 (14.3%) |
| GLRLM (n = 16) | 11 (68.8%) | 1 (6.3%) | 4 (25.0%) |
| GLSZM (n = 16) | 6 (37.5%) | 8 (50.0%) | 2 (12.5%) |
| NGTDM (n = 5) | 3 (60.0%) | 2 (40.0%) | 0 (0.0%) |
| Total (n = 93) | 52 (55.9%) | 31 (33.3%) | 10 (10.8%) |
| Multi-slice Analysis: Resegmentation, fixed bin width = 25, B-spline interpolation | |||
| First order (n = 18) | 3 (16.7%) | 13 (72.2%) | 2 (11.1%) |
| GLCM (n = 24) | 2 (8.3%) | 22 (91.7%) | 0 (0.0%) |
| GLDM (n = 14) | 3 (21.4%) | 11 (78.6%) | 0 (0.0%) |
| GLRLM (n = 16) | 3 (18.8%) | 13 (81.3%) | 0 (0.0%) |
| GLSZM (n = 16) | 3 (18.8%) | 13 (81.3%) | 0 (0.0%) |
| NGTDM (n = 5) | 0 (0.0%) | 5 (100.0%) | 0 (0.0%) |
| Total (n = 93) | 14 (15.1%) | 77 (82.8%) | 2 (2.2%) |
Definition of ICC used = 2-way mixed-effects model, absolute agreement, single rater intraclass correlation coefficient; n, number of radiomics features.