Adaptive activation functions for skin lesion classification using deep neural networks

Abstract

Skin cancer is considered one of the most common human malignancies, and melanoma is the deadliest form of this disease. Early detection can influence the outcome of the disease and improve the chance of surviving. The tremendous improvement of deep learning algorithms in image recognition tasks promises a great success for medical image analysis, in particular, skin lesion classification for melanoma diagnosis. Activation functions play an important role in the performance of deep neural networks for image recognition problems as well as medical image classification. In this paper, we show that a deep neural network model with adaptive piecewise linear units can achieve excellent results in skin disease recognition. Experimental results show that a convolutional neural network model with adaptive piecewise linear units outperforms the same network with different activation functions in the skin lesion classification task. All experiments are performed using the data provided in International Skin Imaging Collaboration (ISIC) 2018 Skin Lesion Analysis towards Melanoma Detection. © 2018 IEEE.