Combining deep learning and hand-crafted features for skin lesion classification

Melanoma is one of the most lethal forms of skin cancer. It occurs on the skin surface and develops from cells known as melanocytes. The same cells are also responsible for benign lesions commonly known as moles, which are visually similar to melanoma in its early stage. If melanoma is treated correctly, it is very often curable. Currently, much research is concentrated on the automated recognition of melanomas. In this paper, we propose an automated melanoma recognition system, which is based on deep learning method combined with so called hand-crafted RSurf features and Local Binary Patterns. The experimental evaluation on a large publicly available dataset demonstrates high classification accuracy, sensitivity, and specificity of our proposed approach when it is compared with other classifiers on the same dataset.