Table 3.
Optimized DCNN architecture determined by Bayesian optimization
| Layer | Test dataset A | Test dataset B | Test dataset C |
|---|---|---|---|
| Input | 224 × 224 × 5 | 224 × 224 × 5 | 224 × 224 × 5 |
| Conv. + ReLu | Filter size: 7 × 7 Num. of filter: 193 | Filter size: 7 × 7x Num. of filter: 214 | Filter size: 7 × 7 Num. of filter: 160 |
| Normalization | Size of the channel window: 5 | Size of the channel window: 5 | Size of the channel window: 5 |
| Max-pooling | Filter size: 3 × 3 | Filter size: 3 × 3 | Filter size: 3 × 3 |
| Conv. + ReLu | Filter size: 5 × 5 Num. of filter: 49 | Filter size: 5 × 5 Num. of filter: 442 | Filter size: Num. of filter: 148 |
| Normalization | Size of the channel window: 5 | Size of the channel window: 5 | Size of the channel window: 5 |
| Max-pooling | Filter size: 3 × 3 | Filter size: 3 × 3 | Filter size: 3 × 3 |
| Conv. + ReLu | Filter size: 5 × 5 Num. of filter: 440 | Filter size: 5 × 5 Num. of filter: 533 | Filter size: Num. of filter: 122 |
| Conv. + ReLu | Filter size: 5 × 5 Num. of filter: 440 | Filter size: 5 × 5 Num. of filter: 533 | Filter size: Num. of filter: 122 |
| Conv. + ReLu | Filter size: 5 × 5 Num. of filter: 440 | Filter size: 5 × 5 Num. of filter: 533 | Filter size: Num. of filter: 122 |
| Max-pooling | Filter size: 3 × 3 | Filter size: 3 × 3 | Filter size: 3 × 3 |
| Conv. + ReLu | Filter size: 3 × 3 Num. of filter: 123 | Filter size: 3 × 3 Num. of filter: 96 | Filter size: 3 × 3 Num. of filter: 209 |
| Conv. + ReLu | Filter size: 3 × 3 Num. of filter: 134 | Filter size: 3 × 3 Num. of filter: 104 | Filter size: 3 × 3 Num. of filter: 228 |
| Conv. + ReLu | Filter size: 3 × 3 Num. of filter: 146 | Filter size: 3 × 3 Num. of filter: 113 | Filter size: 3 × 3 Num. of filter: 247 |
| Conv. + ReLu | Filter size: 3 × 3 Num. of filter: 157 | Filter size: 3 × 3 Num. of filter: 122 | Filter size: 3 × 3 Num. of filter: 266 |
| FC | 2 | 2 | 2 |
| Output | Benign/Malignant | Benign/Malignant | Benign/Malignant |