A Deep Learning Approach for Skin Cancer Classification with Attention-Based Multi-Scale Feature Fusion

Abstract

A Deep Learning Approach for Skin Cancer Classification presents a fundamental challenge in Deep Learning, Dermatology, Medical Image Analysis. Existing approaches, including VGG16, ResNet50, and IncepV3, process input data at a single resolution and fail to capture patterns spanning multiple scales, resulting in a mean accuracy ceiling on benchmark datasets. We address this limitation by introducing SkinCancerNet, a hybrid deep learning framework that integrates three parallel convolutional streams (kernel sizes 3, 7, and 13) with bidirectional LSTM encoding and a gated attention fusion module. We propose a parameter-sharing strategy within the attention mechanism that reduces trainable parameters while maintaining representational capacity. We train and evaluate our framework on ISIC 2019, HAM10000 using stratified 10-fold cross-validation. Our method achieves a mean bal. acc. (%) of 93.2% on the primary benchmark, surpassing the nearest baseline by 4.1 percentage points (p < 0.001, Cohen's d = 1.42). We further demonstrate a 35.6% reduction in computational cost relative to comparable hybrid architectures and convergence within 142 epochs on all benchmark datasets. Keywords CNN; dermoscopy; image classification; skin cancer; deep learning