Supplements for manuscript: Automated classification of hidradenitis suppurativa disease severity by convolutional neural network analyses using clinical images – a prospective study.
Description
Figure legends: Supplemental Fig 1. Study workflow and CNN architecture: The dynamics of the disease were estimated using method A (mixed neural network) and method B with its segmentation variants were implemented to predict the severity scores and localization of lesion area, respectively. Supplemental Fig 2. Individual Typology Angle (ITA°) map. According to their ITA°, individuals were assigned to one of seven skin classification categories. To see if the difference between actual Fitzpatrick (Expert) and ITA phototype (Scarletred®Vision) are significant, a Mann Whitney U Test was conducted (n=149, significance level 5%). The null hypothesis for this test states that there will be no significant difference between the two groups. The test resulted in a p-value of ≈ 0.09 (> 0.05). Therefore, there is no significant difference between the two groups which supports Scarletred®Vision robustness to compute the ITA° related phototype. Supplemental Fig 3. Images were signal augmented and analyzed by measuring SEV*, L*, +a* and +b* values. Clear trends were observed in the SEV*_meandt-refdt and +a*_meandt-refdt (dark green squares, p<0.001), indicating the importance of the SEV* for classification and lesion detection also in this study 13. Supplemental Fig 4. The CNN predicted score aligned with changes of the IHS4, reflecting HS specific dynamics.