CAViT-Net: A Hybrid Deep Learning Model for Accurate and Efficient Skin Disease Detection

Skin disease detection is a critical area in medical diagnostics, requiring robust and accurate methods for early identification and classification of dermatological conditions. This study introduces CAViT-Net, a hybrid deep learning model that combines Convolutional Neural Networks (CNNs) for local feature extraction and Vision Transformers (ViTs) for global spatial relationship learning. The architecture further enhanced with self-attention mechanisms to dynamically focus on critical lesion areas, ensuring high diagnostic precision. CAViT-Net was trained and tested on a curated dataset of skin lesion images, achieving a classification accuracy of 99.2%, with an average precision of 98.9% and F1-score of 97.6% across 10 dermatological conditions, including melanoma, squamous cell carcinoma, and vascular lesions. Comparative analysis demonstrates that CAViT-Net outperforms existing state-of-theart models such as ResNet-50, DenseNet-121, and Vision Transformers, which achieved accuracies of 96.5%, 97.4%, and 94.3%, respectively. Additionally, computational efficiency metrics show that CAViT-Net achieves competitive inference times and memory usage, making it suitable for real-world applications. These results highlight the potential of CAViT-Net to revolutionize automated skin disease diagnostics, offering clinicians a reliable and interpretable tool for decision-making.