Table 4. Classification of images with the gold standard and deep learning framework according to the severity of ulcerations regardless their localization.
Deep learning framework | ||||||
Normal | Aphthae | Ulcer | Fissure/large ulcer | Total | ||
Gold standard | Normal | 885 | 0 | 1 | 1 | 887 |
Aphtae | 9 | 146 | 14 | 0 | 169 | |
Ulcer | 14 | 56 | 65 | 44 | 179 | |
Fissure/large ulcer | 11 | 9 | 39 | 60 | 119 | |
Total | 919 | 211 | 119 | 105 | 1354 |
Data are shown for the patient split used for training. The inter-modality agreement for severity of lesions is substantial (κ = 0.72).