Deep learning techniques applied to skin lesion classification

Skin cancer is a global health issue, and its early diagnosis is essential to avoid mortality. Automated classification of skin lesions can offer a fast and efficient way to diagnose skin lesions accurately. This paper presents an approach to skin lesion classification using deep learning techniques. Our methodology involved a combination of image preprocessing methods, data augmentation, transfer learning, and an ensemble strategy to classify between three skin lesion classes. Our final proposal consisted of the ensemble of EfficientNet-B2, EfficientNet-B5, and ResNeSt101 models. Our model demonstrated high performance on the ISIC 2017 dataset, obtaining a performance of 0.9300 and 0.8132 for AUC and balanced accuracy, respectively.