A multi-tree genetic programming representation for melanoma detection using local and global features

© Springer Nature Switzerland AG 2018. Melanoma is the deadliest type of skin cancer that accounts for nearly 75% of deaths associated with it. However, survival rate is high, if diagnosed at an early stage. This study develops a novel classification approach to melanoma detection using a multi-tree genetic programming (GP) method. Existing approaches have employed various feature extraction methods to extract features from skin cancer images, where these different types of features are used individually for skin cancer image classification. However they remain unable to use all these features together in a meaningful way to achieve performance gains. In this work, Local Binary Pattern is used to extract local information from gray and color images. Moreover, to capture the global information, color variation among the lesion and skin regions, and geometrical border shape features are extracted. Genetic operators such as crossover and mutation are designed accordingly to fit the objectives of our proposed method. The performance of the proposed method is assessed using two skin image datasets and compared with six commonly used classification algorithms as well as the single tree GP method. The results show that the proposed method significantly outperformed all these classification methods. Being interpretable, this method may help dermatologist identify prominent skin image features, specific to a type of skin cancer.