TABLE V. Performance of Networks on LIDC.
| Network | Parameters | ROC-AUC | AP | NLL |
|---|---|---|---|---|
| Dey et al. [6] | - | 0.955 | - | - |
| Kaung et al. [26] | - | 0.943 | - | - |
| Safta et al. [27] | - | 0.977 | - | - |
| CNN-3D | 44.4M | 0.914 | 0.884 | 0.422 |
| CRN1-R | 16.4M | 0.900 | 0.859 | 0.447 |
| CRN1-2D | 16.4M | 0.939 | 0.897 | 0.317 |
| CAN1-R | 14.9M | 0.940 | 0.929 | 0.305 |
| CAN1-2D | 14.9M | 0.950 | 0.933 | 0.299 |
Results for single-time-point classification are shown against competing techniques when separately trained and tested on the LIDC dataset (results from proposed network in bold). We see similar gains in performance with the proposed approach, CAN1-2D, as on the NLSTx dataset.