An image-based segmentation recommender using crowdsourcing and transfer learning for skin lesion extraction

Deep learning is widely used in medical applications regarding the high performance it can achieve. In this paper, we propose a segmentation recommender based on crowdsourcing and transfer learning for skin lesion extraction. In fact, after collecting and pre-processing data from the ISIC2017 segmentation challenge, we tested two pre-trained architectures (VGG16 and ResNet50) to extract features from the convolutional parts. Then, a classifier with an output layer, composed of five nodes representing the segmentation methods' classes, was built. Thus, the proposed architecture is able to dynamically predict the most appropriate segmentation technique for the detection of skin lesions in any input image. Experimental results prove the capability of the proposed image-based method to improve the segmentation performance comparatively to the state of the art methods. (C) 2018 Elsevier Ltd. All rights reserved.