Table 1.
Training details for the deep neural networks.
| Resolution limiting factor | Tissue type | Low resolution input type | Training dataset size (number of patches before augmentation) | Training time (s) | Number of iterations |
|---|---|---|---|---|---|
| Pixel size-limited (System A) | Pap smear | 1 × 1 raw hologram | 56250 | 9,078 | 17,000 |
| Lung | 1 × 1 raw hologram | 83700 | 17,052 | 28,000 | |
| Lung | 2 × 2 raw holograms | 83700 | 9,363 | 15,000 | |
| Lung | 3 × 3 raw holograms | 83700 | 30,480 | 52,500 | |
| Diffraction-limited (System B) | Pap smear | 4×/0.13 NA objective lens | 65475 | 46,411 | 100,000 |
All the networks were trained with a batch size of 10 using 128 × 128 pixel patches.