Author Affiliations
School of Artificial Intelligence and Computer Science, Jiangnan University, Wuxi, Jiangsu 214122, Chinashow less
Fig. 1. Structure of deeply-supervised net
Fig. 2. Flow chart of our experiment
Fig. 3. Network structure proposed in this paper
Fig. 4. Attention module diagrams. (a) Channel attention module; (b) spatial attention module
Fig. 5. Comparison before and after pretreatment. (a) Transverse plane; (b) sagittal plane; (c) coronal plane; (d) HU distribution before pretreatment; (e) HU distribution after pretreatment
Fig. 6. Segmentation results of different methods. (a) Raw image; (b) Ground truth; (c) proposed method; (d) U-Net; (e) U-Net+deeply-supervised net; (f) U-Net+deeply-supervised net+spatial attention
Fig. 7. Comparison between proposed method and One-stage
Input solution | Voxel spacing /mm | Slice | Stride | DSC |
---|
128×128×32 | 3 | 10 | 5 | 0.913 | 128×128×32 | 3 | 15 | 3 | 0.921 | 256×256×32 | 3 | 15 | 3 | 0.944 | 256×256×48 | 2 | 15 | 3 | 0.957 |
|
Table 1. Segmentation results at different input sizes
Model | Liver | Tumor |
---|
DSC | VOE | RVD | DSC | VOE | RVD |
---|
U-Net | 0.939 | 0.112 | 0.007 | 0.547 | 0.411 | -0.070 | U-Net+deeply-supervised net | 0.952 | 0.089 | 0.001 | 0.589 | 0.390 | -0.104 | U-Net+deeply-supervised net+spatial attention | 0.955 | 0.084 | -0.006 | 0.643 | 0.375 | -0.091 | U-Net+deeply-supervised net+FF | 0.957 | 0.081 | 0.003 | 0.676 | 0.341 | -0.064 |
|
Table 2. Comparison of segmentation results of various network structures
Model | Liver | Tumor |
---|
DSC | VOE | RVD | DSC | VOE | RVD |
---|
Proposed | 0.957 | 0.081 | 0.003 | 0.676 | 0.341 | -0.064 | One-stage | 0.944 | 0.114 | 0.017 | 0.585 | 0.380 | -0.081 |
|
Table 3. Comparison between proposed method and One-stage
Model | Liver | Tumor |
---|
DSC | VOE | RVD | DSC | VOE | RVD |
---|
Bi, et al | 0.959 | -- | -- | 0.500 | -- | -- | MEDDIIR | 0.950 | 0.094 | 0.047 | 0.658 | 0.380 | -0.12 | Kaluva, et al[6] | 0.912 | 0.150 | -0.008 | 0.492 | 0.411 | 19.70 | Jin, et al[22] | 0.961 | 0.074 | 0.002 | 0.595 | 0.389 | -0.152 | Chen, et al[23] | -- | -- | -- | 0.650 | -- | -- | Jiang, et al[23] | 0.953 | -- | -- | 0.668 | 0.135 | 0.012 | Our method | 0.957 | 0.081 | 0.003 | 0.676 | 0.341 | -0.064 |
|
Table 4. Comparison of different segmentation methods