Ordinal regression analyses in the test set (n=200)
| Shape-based TR | Texture-based TR | ||||||
| Maximum 2D diameter slice | Voxel volume | Major axis length | Inverse difference moment normalized | Dependence variance | Large dependence emphasis | ||
| Description | Largest Euclidean distance between 2D surfaces of the thrombus mesh vertices per 10 mm | Thrombus volume measured by summing over voxel volumes per 0.1 mL | Major axis length of an ellipsoid fit to the thrombus per 10 mm | Local homogeneity: probability of similar local intensities per 0.01. Higher values indicate more locally homogeneous tissue | Variance over local intensities per 10. Higher values indicate more heterogeneity | Distributions of large regions with similar intensities per 100. Higher values indicate more homogeneity | |
| Feature importance | 0.013 | 0.012 | 0.011 | 0.017 | 0.013 | 0.010 | |
| eTICI | cOR | 0.97 (0.76 to 1.04) | 0.90 (0.74 to 1.12) | 0.90 (0.68 to 1.20) | 0.87 (0.74 to 1.01) | 0.87 (0.67 to 1.11) | 0.84 (0.54 to 1.28) |
| acOR | 0.97 (0.75 to 1.27) | 0.89 (0.70 to 1.12) | 0.86 (0.63 to 1.18) |
0.85
(0.72 to 0.99)* | 0.87 (0.67 to 1.13) | 0.86 (0.55 to 1.36) | |
| Number of attempts to eTICI ≥2B | cOR | 1.07 (0.81 to 1.41) |
1.27
(1.00 to 1.67)* | 1.19 (0.89 to 1.60) |
1.20
(1.03 to 1.43)* |
1.33
(1.03 to 1.73)* |
1.62
(1.05 to 2.51)* |
| acOR | 1.07 (0.79 to 1.47) |
1.36
(1.03 to 1.88)* | 1.24 (0.90 to 1.73) |
1.24
(1.04 to 1.49)* | 1.32 (1.00 to 1.75) | 1.46 (0.92 to 2.34) | |
| 90-day mRS | cOR | 0.87 (0.68 to 1.08) |
0.75
(0.58 to 0.94)* | 0.87 (0.65 to 1.15) | 1.03 (0.89 to 1.18) | 1.05 (0.82 to 1.35) | 1.35 (0.91 to 2.02) |
| acOR | 0.93 (0.71 to 1.21) | 0.83 (0.62 to 1.08) | 0.96 (0.70 to 1.30) | 1.11 (0.95 to 1.30) | 1.07 (0.82 to 1.40) | 1.44 (0.94 to 2.20) | |
Results from ordinal logistic regression models are presented as (adjusted) common odds ratios (acOR) with 95% CI. acOR values were computed on the original data without median IQR normalization. eTICI and 90-day mRS: acOR for a 1-step shift on eTICI scale or 90-day mRS scale, respectively, OR >1 is favorable. Attempts to success: endovascular treatment attempts to achieve TICI ≥2B or <1 is favorable.
*p<0.05. Descriptions of the TR features are available on the PyRadiomics website: https://pyradiomics.readthedocs.io/en/latest/features.html
eTICI, extended Thrombolysis in Cerebral Ischemia; mRS, modified Rankin Scale; TR, thrombus radiomics.