Table 3 Performance of the models against the reference contour in terms of the generalized conformity index (\(\hbox {CI}_{\mathrm{gen}}\)) (mean ± SD) for the native and contrast-enhanced CT datasets.
From: Deep learning-based segmentation of the thorax in mouse micro-CT scans
Algorithm | Models | Native CT (test set 1) | Contrast-enhanced CT (test set 2) | ||||
|---|---|---|---|---|---|---|---|
Heart | Total lungs | Spinal Cord | Heart | Total lungs | Spinal Cord | ||
nnU-Net | 3d_fullres | 90% ± 1% | 95% ± 1% | 65% ± 6% | 87% ± 3% | 92% ± 2% | 75% ± 7% |
2d | 89% ± 2% | 94% ± 1% | 66% ± 6% | 72% ± 22% | 81% ± 16% | 65% ± 14% | |
AIMOS | UNet-768 | 89% ± 2% | 93% ± 1% | 64% ± 6% | 74% ± 18% | 76% ± 18% | 73% ± 8% |
Interobserver variability | 79% ± 3% | 86% ± 2% | 60% ± 6% | 80% ± 3% | 81% ± 8% | 69% ± 6% | |
