Pigmented Skin Lesions Classification using Convolutional Neural Networks

In this paper we present an architecture for classification of pigmented skin lesions from dermatoscopic images. The architecture is using image pre-processing for natural hair removal and image segmentation for extraction of the skin lesion area. The segmented images were processed by a convolutional neural network classifier. The training process was done by using the Keras and TensorFlow python packets with CUDA supported. The best performance was achieved by a convolutional neural network architecture with three convolution layers and the classification accuracy was equal to 76.83%.