Table 2.
Performance metrics of AI methods on stereological annotations, grouped by slab.
| Slab n. | Tot. markers | Method | Prec. (%) | Rec. (%) | (%) |
|---|---|---|---|---|---|
| 6 | 379 | BCFind-v2 | 69.0 | 84.4 | 75.9 |
| StarDist (ResNet) | 74.7 | 79.7 | 77.1 | ||
| StarDist (UNet) | 81.5 | 70.7 | 75.7 | ||
| CellPose | 81.5 | 36.2 | 50.1 | ||
| 18 | 746 | BCFind-v2 | 76.9 | 68.7 | 72.6 |
| StarDist (ResNet) | 76.5 | 61.8 | 69.6 | ||
| StarDist (UNet) | 81.2 | 39.9 | 53.5 | ||
| CellPose | 78.9 | 47.0 | 58.9 | ||
| 30 | 626 | BCFind-v2 | 72.0 | 81.6 | 76.5 |
| StarDist (ResNet) | 73.6 | 81.8 | 77.5 | ||
| StarDist (UNet) | 77.0 | 81.9 | 79.4 | ||
| CellPose | 76.1 | 56.4 | 64.8 | ||
| 42 | 494 | BCFind-v2 | 76.4 | 82.6 | 79.4 |
| StarDist (ResNet) | 74.8 | 84.6 | 79.4 | ||
| StarDist (UNet) | 78.0 | 81.2 | 79.6 | ||
| CellPose | 83.8 | 47.2 | 60.4 | ||
| Tot. | 2245 | BCFind-v2 | 73.8 | 78.0 | 75.8 |
| StarDist (ResNet) | 75.6 | 75.4 | 75.5 | ||
| StarDist (UNet) | 78.9 | 65.9 | 71.8 | ||
| CellPose | 79.3 | 47.8 | 59.7 |
DL models are here trained on volumes from slabs 1–5. Bold values are the highest results per metric and considered set within a maximum distance of 0.5 point percentage to the best model.