Table 2 Results of feature selection for each dataset
From: Are radiomics features universally applicable to different organs?
| Dataset | Feature category | Feature name | Repetition times |
|---|---|---|---|
| Training (7 features) | Histogram-based | 2.5 percentile | 20 |
| 97.5 percentile | 20 | ||
| Shape-based (3D) | Sphericity | 20 | |
| Max 3D diameter | 20 | ||
| Shape-based (2D) | Roundness factor | 20 | |
| Texture-based | Informational Measure of Correlation, subsampled GLCM | 17 | |
| Size zone variability | 10 | ||
|
Test set1 (6 features) | Histogram-based | Variance | 20 |
| 50 Percentile | 20 | ||
| Texture-based | Homogeneity, subsampled GLCM | 15 | |
| Shape-based (2D) | Roundness factor | 9 | |
| Fractal-based | FSD | 4 | |
| Histogram-based | Standard deviation | 1 | |
|
Test set2 (3 features) | Texture-based | Energy, subsampled GLCM | 1 |
| Contrast, NGTDM | 1 | ||
| Difference entropy, subsampled GLCM | 1 | ||
|
Test set3 (2 features) | Texture-based | Informational Measure of Correlation, subsampled GLCM | 19 |
| Coarseness, NGTDM | 3 |