Enhancing Skin Cancer Detection Using Hybrid Deep Neural Network (HDNN) Approach

Abstract

Skin cancer remains one of the most aggressive and potentially fatal forms of cancer worldwide. Its early detection is paramount, as timely diagnosis significantly enhances the likelihood of successful treatment and long-term survival. However, traditional diagnostic methods often require expert dermatological assessment, which may not always be readily available, especially in resource-constrained environments. To address this challenge, we present a novel skin cancer detection framework based on a Hybrid Deep Neural Network (HDNN) architecture. The proposed model integrates the strengths of multiple deep learning paradigms to achieve robust feature extraction and classification. It was rigorously evaluated using a publicly available dermoscopic image dataset and achieved a maximum classification accuracy of 87.33% on the test set. In addition to presenting the architecture and performance of the HDNN, we conducted a comparative analysis with several state-of-the-art models, including DenseNet201, EfficientNetB0, MobileNet, and ResNet152. The results demonstrate that our HDNN model outperforms these baselines in terms of both accuracy and consistency. The improved diagnostic precision achieved through our approach has significant implications for clinical practice, as it can enable faster, more reliable screening of skin lesions. Ultimately, this could contribute to earlier interventions, reduced treatment costs, and improved patient outcomes. Our findings highlight the potential of deep learning in advancing skin cancer diagnostics.