Authors: Youssef Filali; My Abdelouahed Sabri; Abdellah Aarab
Addresses: Department of Computer Science, Faculty of Sciences Dhar-Mahraz, USMBA, Fez, Morocco ' Department of Computer Science, Faculty of Sciences Dhar-Mahraz, USMBA, Fez, Morocco ' Department of Physics, Faculty of Sciences Dhar-Mahraz, USMBA, Fez, Morocco
Abstract: Skin cancer is one of the most threatening cancer all over the world. The early detection of this skin cancer could help dermatologists in saving patients' lives. For that, a computer-aided diagnosis is used for an early evaluation of this kind of cancer. Skeletons of the lesions are an effective representation making it possible to properly describe the shape and size of lesions and thus used to classify them effectively as melanoma or non-melanoma. Therefore, the proposed idea in this paper is to use the lesions skeleton as deep learning entry instead of the original images. Experimentation shows that this idea can both increase the classification rate, in comparison with recent approaches from the literature, and thus reduce the number of layers used to create the deep network. The accuracy of our proposed approach on the well-known ISIC challenge and dermoscopy datasets is 95%, showing the effectiveness of our system.
Keywords: skin cancer; melanoma; deep learning; convolutional neural network; CNN; skeleton.
International Journal of Cloud Computing, 2021 Vol.10 No.5/6, pp.565 - 578
Received: 05 Jan 2020
Accepted: 12 Mar 2020
Published online: 10 Jan 2022 *