Skin Lesions Classification from Digital Images using Deep Convolutional Neural Network

Abstract

Skin cancer is one of the most commonly diagnosed cancers worldwide, and its early detection is essential for effective treatment and improved patient survival rates. Conventional diagnostic methods often rely on manual visual inspection by dermatologists, which may result in subjective interpretations and diagnostic inaccuracies. To address these limitations, this study proposes an enhanced DenseNet-based Convolutional Neural Network (CNN) model integrated with a Channel Refinement Module (CRM) for efficient multiclass skin lesion classification. newlineThe proposed hybrid CRM-DenseNet model was trained and validated using a comprehensive dataset consisting of six categories of skin cancer lesions, including melanoma, basal cell carcinoma, squamous cell carcinoma, actinic keratosis, benign keratosis, and melanocytic nevus. The integration of the CRM improved the model s ability to extract and refine essential feature maps, leading to superior classification accuracy compared to conventional CNN architectures. newlineExperimental evaluation demonstrated that the proposed model achieved an accuracy of approximately 98.14% and a Cohen Kappa score of 0.95, indicating an excellent level of agreement and reliability. The results clearly established that the CRM-DenseNet model outperforms existing architectures in terms of precision, recall, and F1-score, making it highly effective for real-world dermatological applications. newlineThis study concludes that the CRM-DenseNet model provides a robust, reliable, and scalable solution for the early detection and diagnosis of skin cancer. Its adaptability also suggests potential for integration into portable diagnostic tools and clinical systems to facilitate real-time, accurate, and cost-effective skin cancer screening. newline newline