2021/04/20 by Alceu Bissoto, Eduardo Valle, Bissoto, Alceu +3 · 4 citations
Computer Science · Medicine · #AI in cancer detection #Computer Vision and Pattern Recognition (cs.CV) #Cutaneous Melanoma Detection and Management #FOS: Computer and information sciences #FOS: Electrical engineering #Genital Health and Disease #Image and Video Processing (eess.IV) #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.2104.10603
openalex publication_date 2021/04/20 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Despite the growing availability of high-quality public datasets, the lack of\ntraining samples is still one of the main challenges of deep-learning for skin\nlesion analysis. Generative Adversarial Networks (GANs) appear as an enticing\nalternative to alleviate the issue, by synthesizing samples indistinguishable\nfrom real images, with a plethora of works employing them for medical\napplications. Nevertheless, carefully designed experiments for skin-lesion\ndiagnosis with GAN-based data augmentation show favorable results only on\nout-of-distribution test sets. For GAN-based data anonymization - where the\nsynthetic images replace the real ones - favorable results also only appear\nfor out-of-distribution test sets. Because of the costs and risks associated\nwith GAN usage, those results suggest caution in their adoption for medical\napplications.\n