Joint Dermatological Lesion Classification and Confidence Modeling with Uncertainty Estimation

Deep learning has played a major role in the interpretation of dermoscopic images for detecting skin defects and abnormalities. However, current deep solutions dermatological lesion analysis are typically limited providing probabilistic predictions which highlights importance concerning uncertainties. This concept uncertainty can provide confidence level each feature prevents overconfident with poor generalization on unseen data. In this paper, we propose an overall framework that jointly considers classification estimation together. The estimated to avoid uncertain undesirable shift, caused by environmental difference input image, latent space is pooled from network. Our qualitative results show modeling uncertainties not only helps quantify model prediction but also layers focus confident features, therefore, improving accuracy classification. We demonstrate potential proposed approach two state-of-the-art datasets (ISIC 2018 ISIC 2019).