Accurate classification of nail diseases like psoriasis and onychomycosis is still a clinically important issue owing to inter-class visual similarity and fine morphological differences. This paper introduces a new deep learning structure, Dermatology Neural Network (DERMANet), which is developed for the classification of nail diseases automatically. The new approach synergistically combines the ConvNeXt backbone with a Convolutional Block Attention Module (CBAM) to reinforce spatial and channel-wise feature recalibration to sharpen model attention on diagnostically important regions. The model was evaluated using a standard nail disease dataset containing three categories: healthy, psoriasis, and onychomycosis. Comparative experiments were carried out against Convolutional Neural Network, ResNet 50, VGG 16, Vision Transformer and vanilla ConvNeXt under identical training scenarios. Out of all the architectures tested, the proposed model DERMANet had the highest classification accuracy of 97% with the best discriminative capability and generalization performance. Empirical findings confirm the effectiveness of combining attention mechanisms with state-of-the-art convolutional architectures for accurate and robust dermatoscopic image analysis. The proposed model is promising for use in computer-aided dermatology diagnostics, supporting quick and objective evaluation of nail pathologies. • A new deep learning model, DERMANet, is proposed for automatic nail disease classification. • ConvNeXt backbone is integrated with a CBAM attention module to enhance feature focus. • The model achieves 97% accuracy on a standard three-class nail disease dataset. • DERMANet outperforms CNN, ResNet50, VGG16, ViT, and vanilla ConvNeXt under identical training. • The approach shows strong generalization and potential for computer-aided dermatology diagnostics.
Bhuvanya et al. (Sun,) studied this question.