Enhancing nail disease diagnosis: a capsule network with SE attention and dual backbone models.

Nail diseases, including fungal infections and malignancies, pose significant health risks and may lead to severe complications if not accurately diagnosed. Conventional diagnostic methods are often subjective and time-consuming. This study introduces CapsuleSEDualNet, a novel deep learning framework designed to achieve robust and interpretable multi-class nail disease diagnosis. The proposed CapsuleSEDualNet integrates a Capsule Network head with a Squeeze-and-Excitation (SE) attention mechanism within a Dual-Backbone architecture combining MobileNetV2 and DenseNet121. The SE block enhances feature discriminability, while the Capsule head preserves spatial hierarchies for improved interpretability. To address dataset imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied. The model was evaluated using extensive experiments and benchmarked against existing deep learning architectures. CapsuleSEDualNet achieved an overall classification accuracy of 96%, demonstrating superior performance compared to baseline models. The integration of SE attention and Capsule networks effectively reduced misclassification rates and improved feature representation. Experimental findings confirm the framework's robustness, scalability, and interpretability for clinical applications. The proposed CapsuleSEDualNet framework provides an efficient and reliable solution for automated nail disease screening. By combining diagnostic accuracy, model interpretability, and computational efficiency, it addresses a critical clinical need for early and accessible dermatological assessment. The model shows strong potential as a computer-aided decision support tool, pending further clinical validation. Its potential integration into computer-aided dermatology systems and telemedicine platforms can enhance physician decision-making, reduce misdiagnosis risk, and improve patient outcomes in dermatological care.