Deep Learning Diagnosis of Pigmented Skin Lesions

Convolutional Neural Networks allows prediction and automation of various tasks in interdisciplinary fields. These models have the ability to learn various features from image representations. These also require a substantial amount of observations for accurate classification. This paper presents the diagnosis of skin lesion types using deep learning models. We used various methods -involving deep learning, like parallel networks, transfer learning and CNNs with segmentation. For this work, we used HAM10000 dataset which contains labeled images of skin lesions. Various pre-trained CNN models were fine-tuned to predict skin lesion types. We report the best accuracy of 82.8% and an average F-score of 0.7.