Skin Cancer Detection Using Semi-Unsupervised Deep Learning in Generative Adversarial Networks: A Study on Melanoma Detection

Melanoma is widely recognized as the most life-threatening type of skin cancer, and early detection plays a critical role in improving a patient's chances of survival. While various deep learning techniques, including object detection and supervised learning utilizing frameworks such as Visual Geometry Group Neural Network (VGGNet) and Residual Neural Network (ResNet), have been employed for skin cancer detection, this study proposes the use of Deep Learning with Deep Convolutional Generative Adversarial Networks (DCGANs) to address the challenge of melanoma detection. The presented experiment achieved an accuracy of 88.33% using a model with only 9 layers and less than 700 thousand parameters, resulting in a parameter size of 2.55 MB and an estimated total size of 29.16 MB. The approach resulted in a substantial increase in both the accuracy and speed of the model, demonstrating the effectiveness of the proposed method. This methodology aligns with practical scenarios where labeling all melanoma objects by humans may not be feasible.