Table 2 3DIRCAD database: performance of the proposed method using five configurations of training data at the processing stage and three configurations of the segmentation system
From: Three-stage segmentation of lung region from CT images using deep neural networks
Training data | Dice score | |||
|---|---|---|---|---|
Training images | Training labels | Three-stage system | Without CNNs | Without contrast enhancement |
Ground truth images | Ground truth images | 0.81 | 0.76 | 0.79 |
Otsu-based binary images | Otsu-based binary images | 0.85 | 0.81 | 0.80 |
Foreground images | Foreground images | 0.83 | 0.77 | 0.80 |
Grayscale images | Two-class k-means clustered images | 0.90 | 0.81 | 0.80 |
Grayscale images | Three-class k-means clustered images | 0.92 | 0.83 | 0.82 |