Enabled Deep CNN Model for Skin Cancer Classification Using Dermoscopic Images

Skin cancer is identified as one of the most common and dangerous dermatological conditions, and it is essential that it is diagnosed early and correctly to significantly increase patient survival rates. Although dermoscopic image analysis with deep learning has been proven effective for dermoscopic image analysis, several problems like non-relevant background characteristics, poor feature extraction, and lack of explanations still exist. In light of these limitations, this paper presents an attention-assisted hybrid deep learning framework for dermoscopic image classification tasks to detect and classify skin cancer based on dermoscopic images. The proposed deep learning framework combines EfficientNet-B4 and CBAM for deep feature extraction and lesion-associated channel and spatial cues. Moreover, GWO is utilized for the optimization of features and parameters, which further leads to faster convergence and generalization capabilities. The proposed framework is evaluated on a publicly available dermoscopic image dataset, and results show that it outperforms existing deep learning and non-attention models with higher classification accuracy at 98.7 percent. The proposed framework is a highly accurate and interpretable system that utilizes deep learning and GWO optimization and is efficient and practical for real-world application in dermoscopic image analysis tasks like computer-assisted diagnosis for skin cancer.