Proposing a novel Cascade Ensemble Super Resolution Generative Adversarial Network (CESR-GAN) method for the reconstruction of super-resolution skin lesion images

Skin cancer is one of the most malignant cancers worldwide. Its early detection plays a prominent role in the patients' treatment. The quality of skin lesion images to ease the diagnosis of skin cancer is highly regarded. One of the most common technologies to take the skin lesion images is through a dermoscopy device. However, it is not accessible to all people. Capturing the images via other technologies such as mobile devices, is available everywhere, although they suffer from poor quality. In this paper, a novel Cascade Ensemble Super Resolution Generative Adversarial Network (CESR-GAN) method is proposed to reconstruct super-resolution skin lesion images using low-resolution counterparts. Specifically, a novel feature-based measurement loss function is designed to obtain more details as much as possible and generate higher quality images. Experimental results from quantitative and qualitative comparisons between our CESR-GAN model and other state-of-the-art methods show that our proposed method outperforms the compared methods on ISIC, and PH2 datasets, respectively. The CESR-GANs method can be used to generate super resolution skin images of skin lesions with highly notable performances.