Skip to main content
. 2020 Sep 29;21(10):158–169. doi: 10.1002/acm2.13024

Table 1.

Quantitative evaluations of the automatic segmentation results with DSD‐UNET and 3D U‐Net for the HR‐CTV and OARs (Mean ± Standard deviation)

Metrics Models Regions of interest
HR‐CTV Bladder Small intestine Sigmoid Rectum
DSC (%) DSD‐UNET 82.9 ± 4.1 86.9 ± 3.2 80.3 ± 5.8 64.5 ± 7.9 82.1 ± 5.0
3D U‐Net 74.2 ± 6.2 80.2 ± 4.1 75.1 ± 6.7 55.2 ± 8.6 77.1 ± 6.2
JI (%) DSD‐UNET 72.2 ± 4.3 77.9 ± 3.5 69.4 ± 6.1 52.2 ± 8.3 71.5 ± 5.2
3D U‐Net 60.2 ± 6.5 68.2 ± 4.6 61.9 ± 6.9 42.8 ± 8.8 64.3 ± 6.5
HD (mm) DSD‐UNET 8.1 ± 2.3 12.1 ± 4.0 27.8 ± 10.8 19.6 ± 8.7 9.2 ± 4.6
3D U‐Net 10.5 ± 2.9 16.3 ± 4.2 32.2 ± 11.9 23.4 ± 12.7 12.8 ± 6.0