Abstract
This paper presents a novel strategy that employs Generative Adversarial Networks (GANs) to augment data in the skin lesion segmentation task, which is a fundamental first step in the automated melanoma detection process. The proposed framework generates both skin lesion images and their segmentation masks, making the data augmentation process extremely straightforward. In order to thoroughly analyze how the quality and diversity of synthetic images impact the efficiency of the method, we remodel two different well known GANs: a Deep Convolutional GAN (DCGAN) and a Laplacian GAN (LAPGAN). Experimental results reveal that, by introducing such kind of synthetic data into the training process, the overall accuracy of a state-of-the-art Convolutional/Deconvolutional Neural Network for melanoma skin lesion segmentation is increased.
Highlights
Malignant melanoma is the most dangerous type of skin cancer, with a substantial death rate
In order to employ Generative Adversarial Networks (GANs)-generated images discussed in Section 3, we design a two-step training process
In order to prove the effectiveness of the GANs on the final accuracy, we train a version of each Convolutional-Deconvolutional Neural Network (CDNN) using only real data, for a total of 400 epochs
Summary
The clinical diagnosis for melanoma, achieved with the unaided eye, is slightly better than 60% [12]. A great tool to improve the clinical decision making can be automated analysis through dermoscopic images, which are obtained by a non-invasive in vivo examination with a microscope, exploiting incident light and oil/gel immersion to make subsurface structures of the skin accessible to visual examination This is why the International Skin Imaging Collaboration (ISIC) has begun to aggregate a large-scale, publicly accessible dataset of dermoscopic images (Fig. 1). Experimental results show that adding GAN-generated data in the training process effectively improves the segmentation accuracy of state-of-the-art CDNNs. The rest of the paper is organized as follows: in Section 2 a brief review of the learning strategies exploited in our work is reported.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
