Table 1 Five different training scenarios
From: Melanoma diagnosis using deep learning techniques on dermatoscopic images
Model | Technique to modify the malignant/benign training ratio | Malignant/Benign training ratio |
|---|---|---|
Model 1 | None | 0.231 |
Model 2 | Data augmentation applied to “malignant” class images, no changes to “benign” class images | 0.440 |
Model 3 | Data augmentation applied to “malignant” class images, down sampling applied to “benign” class images | 0.463 |
Model 4 | Data augmentation applied to “malignant” class images, down sampling applied in “benign” class images | 1 |
Model 5 | Data augmentation and addition of PH2 images applied to “malignant” class images, no changes to “benign” class images | 0.256 |