Table 3.
Metrics for AI segmentations without manual adjustments applied to the test cohort (n = 41)
| Pixel-wise | AI model | ||||
|---|---|---|---|---|---|
| Model 1 | Model 2 | Model 3 | Model 4 | Ensemble | |
| Dice | 0.801 (0.206) | 0.817 (0.176) | 0.768 (0.234) | 0.763 (0.233) | 0.801 (0.196) |
| Precision | 0.772 (0.258) | 0.816 (0.223) | 0.752 (0.279) | 0.787 (0.258) | 0.786 (0.250) |
| Sensitivity | 0.893 (0.173) | 0.860 (0.180) | 0.869 (0.182) | 0.821 (0.231)* | 0.872 (0.177) |
| Lesion-wise | |||||
|---|---|---|---|---|---|
| Dice | 0.847 (0.286) | 0.828 (0.264) | 0.809 (0.268) | 0.803 (0.258) | 0.850 (0.278) |
| Sensitivity | 0.854 (0.230) | 0.827 (0.234) | 0.843 (0.228) | 0.831 (0.243) | 0.844 (0.238) |
All values calculated as mean of the 41 patients of the test cohort with standard deviation in parentheses. Bold numbers mark the highest value across the models/ensemble in each evaluation metric. *Denotes statistically significant difference in sensitivity between Model 4 and Model 1 (p = 0.017)