Abstract
Early detection of melanoma, one of the deadliest types of cancer, is of paramount importance. Currently, the use of Convolutional Neural Networks (CNNs) is the main line of investigation for the automated detection of this kind of disease. Most of the existing works, however, were designed based on transferlearning general-purpose architectures to the domain of skin lesions, posing inflexibility and high processing costs to the task. In this work, we introduce a novel architecture that benefits from cutting-edge CNNs techniques Aggregated Transformations combined to the mechanism of Squeeze-and-Excite organized in a residual block; our architecture is designed and trained from scratch having the melanoma problem as goal. Our results demonstrate that such an architecture is competitive to major state-of-the-art architectures adapted to the melanoma detection problem. Having a fraction of the number of weights of previous works, our architecture is prone to evolve and to provide low processing cost for real-world in situ applications.