Table 4.
Results of the automatic segmentation of selected brain structures.
| Dice score (%) | |||
|---|---|---|---|
| Structure | ceT1 | hrT2 | ceT1 + hrT2 |
| VS | 93.9 ± 4.0 | 90.7 ± 5.4 | 94.1 ± 4.1 |
| Pons | 97.6 ± 0.8 | 96.7 ± 0.9 | 97.6 ± 0.8 |
| Brainstem | 96.3 ± 1.3 | 94.4 ± 1.3 | 96.3 ± 1.2 |
| Cerebellar vermal lobules I-V | 93.2 ± 2.3 | 89.3 ± 2.3 | 93.2 ± 2.3 |
| Cerebellar vermal lobules VI-VII | 87.8 ± 5.1 | 84.0 ± 5.1 | 87.8 ± 5.1 |
| Cerebellar vermal lobules VIII-X | 93.1 ± 2.3 | 89.3 ± 2.3 | 93.1 ± 2.3 |
| Right cerebellum | 95.8 ± 1.4 | 93.9 ± 1.5 | 95.8 ± 1.4 |
| Left cerebellum | 95.8 ± 1.5 | 93.8 ± 1.6 | 95.8 ± 1.5 |
The segmentation model trained with nnU-Net was compared to the ground truth segmentations. Inputs are contrast-enhanced T1-weighted (ceT1) images, high-resolution T2-weighted (hrT2) images, or a combination of both (ceT1 + hrT2). The error ranges correspond to the standard deviation of the mean Dice score of all test cases.