Advancing Melanoma Detection using Progressive Growing Generative Adversarial Network: A Novel Generative Approach
نویسندگان
1 Electrical Engineering Section, University Polytechnic, Aligarh Muslim University, Aligarh, India
2 Electrical Engineering Section, University Polytechnic, Aligarh Muslim University, Aligarh, India
3 Electrical Engineering Section, University Polytechnic, Aligarh Muslim University, Aligarh, India
4 Department of Computer Engineering, Zakir Hussain College of Engineering & Technology, Aligarh Muslim University, Aligarh, India
doi
10.5829/ije.2026.39.09c.20چکیده
Melanoma is one of the most deadly types of skin cancer, and finding it early and correctly is very important for helping patients live longer. Traditional diagnostic methods require a lot of resources and expert interpretation, which is why automated solutions are needed. In this research, we introduce TP-GAN, a generative framework that combines Progressive Growing GAN (PGGAN) with a Student’s t-distribution, squeeze-and-excitation blocks, and dynamic residual scaling to produce varied, high-resolution melanoma images. TP-GAN makes synthetic images that are used to balance the ISIC-2020 dataset and teach an Xception classifier through transfer learning. We also made a web-based diagnostic interface that lets us classify uploaded skin lesion images in real time to make it easier to use. Experimental evaluation shows that TP-GAN produces better images (FID = 1.012, IS = 3.249) and, when used with Xception, produces the best diagnostic results (accuracy = 98.25%, sensitivity = 98.02%, specificity = 98.52%, AUC = 0.99), beating other GAN-based methods. By uniting advanced generative modeling, deep learning classification, and user-centered design, this work provides a fast, reliable, and accessible tool for melanoma detection, with potential to support early intervention and improved clinical outcomes.
