Skin Cancer Detection Using Multi-Model Neural Networks

Abstract

Skin cancer is a prevalent and potentially life-threatening condition, with early detection playing a crucial role in improving patient outcomes. This study proposes a novel approach for skin cancer detection by leveraging the complementary strengths of three distinct convolutional neural network architectures: Convolutional Neural Networks (CNN), Visual Geometric Group (VGG16), and Residual Networks (ResNet). Our multi-model neural network system is designed to enhance the overall performance of skin cancer detection by combining the feature extraction capabilities of different architectures. The CNN serves as the baseline model, capturing essential image patterns. VGG16, known for its depth and simplicity, contributes to the network's ability to recognize intricate features. Meanwhile, ResNet, with its residual learning framework, aids in mitigating the vanishing gradient problem and facilitates the training of deeper networks. To evaluate the proposed multi-model approach, we employed a comprehensive dataset comprising diverse skin lesion images, including malignant and benign cases. Transfer learning techniques were utilized to pre-train the models on large-scale image datasets, enhancing their ability to generalize to skin cancer detection. The individual models and the integrated multi-model network were extensively evaluated and compared against traditional methods and single-model architectures. The results demonstrate that the multi-model neural network consistently outperforms individual models and achieves state-of-the-art accuracy in skin cancer detection. The fusion of features extracted by CNN, VGG16, and ResNet enhances the model's ability to discern subtle patterns and improves overall diagnostic accuracy. Additionally, the proposed system exhibits robustness across different skin cancer subtypes and provides interpretable insights into the decision-making process.