Identifying Melanoma in Lesion Images Using Cycle-Consistent Adversarial Networks-Based Data Augmentation

Tao, Mengjun and Yan, Youwei (2021) Identifying Melanoma in Lesion Images Using Cycle-Consistent Adversarial Networks-Based Data Augmentation. In: Proceedings of 2021 International Conference on Medical Imaging and Computer-Aided Diagnosis, MICAD 2021 - Medical Imaging and Computer-Aided Diagnosis :. Lecture Notes in Electrical Engineering, 784 . Springer, Singapore, pp. 21-28. ISBN 9789811638794

Abstract

Early detection of melanoma is extremely important because melanoma is curable at the early stage. Due to the state-of-the-art performance of the Convolutional Neural Networks (CNNs), the CNNs have been widely used for the task. However, hand labeled data is not easily obtained in practical settings. In this paper, we firstly employ generative adversarial network (GAN) to artificially enlarge the dataset, which can generate fake data based on the generative confrontation network. Therefore, the problem of insufficient training samples in melanoma classification tasks has been alleviated. Second, CNNs is employed in our paper to automatically classification, which proved to be more effectively solve the problem of small discrimination between different categories. Based on the proposed method, the experimental results show that the use of deep learning technology can effectively improve the performance of the model in the melanoma classification task, with an average accuracy value of 94.5%, which is nearly 1.9% higher than the previous approaches.