Table 5.
Performance of modified LeNet5 model architecture with a separable convolutional layer as the first layer, and a Type 1 DSCN model with 4 convolutional layers for prediction of BCR and BC survival
| Model | Layer type | Depth | No. params | Batch size | Class label | ROC AUC |
|---|---|---|---|---|---|---|
| Model 1 | S-2D CONV | 4 | 16.7 × 106 | 8 | BCR survival | 0.641 ± 0.095 |
| Model 1 | S-2D CONV | 4 | 16.7 × 106 | 8 | BCR | 0.600 ± 0.122 |
| Model 2 | DSC1 | 6 | 1.3 × 106 | 8 | BCR survival | 0.608 ± 0.068 |
| Model 2 | DSC1 | 6 | 1.3 × 106 | 8 | BCR | 0.628 ± 0.098 |
Here, depth is number of layers with learnable parameters, excluding the output single node layer.
Values in bold denote the highest classification performance achieved for each network architecture and classification task.