Generating Highly Realistic Images of Skin Lesions with GANs

Abstract

As many other machine learning driven medical image analysis tasks, skin image analysis suffers from a chronic lack of labeled data and skewed class distributions, which poses problems for the training of robust and well-generalizing models. The ability to synthesize realistic looking images of skin lesions could act as a reliever for the aforementioned problems. Generative Adversarial Networks (GANs) have been successfully used to synthesize realistically looking medical images, however limited to low resolution, whereas machine learning models for challenging tasks such as skin lesion segmentation or classification benefit from much higher resolution data. In this work, we successfully synthesize realistically looking images of skin lesions with GANs at such high resolution. Therefore, we utilize the concept of progressive growing, which we both quantitatively and qualitatively compare to other GAN architectures such as the DCGAN and the LAPGAN. Our results show that with the help of progressive growing, we can synthesize highly realistic dermoscopic images of skin lesions that even expert dermatologists find hard to distinguish from real ones.

Dermoscopy Image Classification Based on StyleGAN and DenseNet201

“…Baur, Christoph et al [36] utilized the progressive growing concept, which was both quantitatively and qualitatively compared to other GAN architectures such as the DC-GAN and the LAPGAN. Their results showed that with the help of progressive growing, highly realistic dermoscopic images of skin lesions can be synthesized that even expert dermatologists find difficulty distinguishing from real ones.…”

Artificial intelligence and dermatology: opportunities, challenges, and future directions

“…One interesting application is data augmentation, which is the use of a GAN to synthesize real-appearing images of otherwise rare medical conditions in order to train better AI algorithms. This has already been prototyped for synthetic images 39,40 ranging from skin lesions to mammograms 41,42 and echocardiograms. 43 Consider the recently developed DNN for the diagnosis of nonpigmented skin cancer.…”

Mask2Lesion: Mask-Constrained Adversarial Skin Lesion Image Synthesis

“…GANs have also been used to generate various medical imaging modalities, such as generating liver lesion images to augment the CT lesion classification training dataset [6], generating chest X-ray images to augment the dataset for abnormality detection [14], and generating brain CT images from corresponding brain MR images [23]. Skin lesion synthesis tasks have also relied upon GAN-based approaches, such as generating images of benign and malignant skin lesions [3], modeling skin lesions using semantic label maps and superpixels in order to generate new lesion images [4], and generating skin lesions along with their corresponding segmentation masks [17].…”