Skin lesion diagnosis with imbalanced ECOC ensembles

Official URL: https://dx.doi.org/10.1007/978-3-030-64580-9_25

Diagnosis of skin lesions is a challenging task due to the similarities between different lesion types, in terms of appearance, location, and size. We present a deep learning method for skin lesion classification by fine-tuning three pre-trained deep learning architectures (Xception, Inception-ResNet-V2, and NasNetLarge), using the training set provided by ISIC2019 organizers. We combine deep convolutional networks with the Error Correcting Output Codes (ECOC) framework to address the open set classification problem and to deal with the heavily imbalanced dataset of ISIC2019. Experimental results show that the proposed framework achieves promising performance that is comparable with the top results obtained in the ISIC2019 challenge leaderboard.