A Deep Ensemble Feature Extraction and Classification Framework for Melanoma Identification

Skin cancer is a highly severe manifestation of the disease, with the potential to metastasize to other regions of the body if not identified in its early stages. According to WHO and the American Cancer Society (ACS), it is one of the leading causes of mortality in the worldwide population. Early discovery of skin cancer, however, can aid in a proper diagnosis to lessen the disease's effects on people. Dermoscopy Several methods have been introduced to develop automated Computer-Aided Diagnosis (CAD) systems where machine learning based solutions are widely adopted in this domain. To improve the detection accuracy, researchers have introduced deep learning-based solutions. However, the detection performance is affected by several factors, such as the uneven boundary of skin cancer, pigmentation, and hairs on dermoscopy images. Therefore, a novel approach for feature extraction by using handcrafted shape and texture features is introduced in this work. Moreover, the proposed approach adopts pre-trained deep learning architectures for deep feature extraction. The acquired features are combined and subjected to an ensemble classification technique, which uses decision trees, random forests, and support vector machines. A majority voting categorization is used to arrive at the final choice. The outcome of this approach is validated on publically accessible datasets such as PH2, ISIC 2017, and ISIC 2018.