Multi-class Skin Cancer Detection using Deep Learning

Overview

Skin cancer is one of the most common and treatable cancers when detected early. This project develops a deep learning pipeline for multi-class skin cancer classification from dermoscopic images, with a focus on training stability, data augmentation, and model interpretability.

Methodology

• Stochastic Gradient Descent (SGD) Augmentation — novel augmentation strategy using SGD-driven perturbations to improve training robustness on limited medical datasets
• Convolutional Neural Network (CNN) — deep architecture for multi-class dermoscopic image classification across skin lesion types
• Activation Mapping (Grad-CAM) — visual explanations highlighting image regions that drive the classification decision, critical for clinical trust

Key Contributions

• SGD-based augmentation strategy tailored for limited medical imaging datasets
• Multi-class classification across multiple skin cancer types
• Activation mapping provides visually interpretable evidence for clinical review
• Competitive accuracy on benchmark dermoscopic datasets

Publication

Published in Journal of Innovative Image Processing, Vol. 7, No. 4, pp. 1415–1435, 2025.

(Dahat et al., 2025)

Status

Published — JIIP, 2025.