DermaGuard-Net: An Attention-Enhanced Deep Learning Framework for Robust Psoriasis Classification

Abstract

Psoriasis is a long-lasting autoimmune disease of the skin, which affects more than 125 million individuals worldwide and is marked by inflamed and scaled skin lesions and major comorbidities. The conventional methodologies of diagnosis namely clinical examination, histological and dermoscopic imaging are biased, time consuming and rely on dermatological experience. Artificial intelligence (AI) and especially deep learning (DL) is a game-changer in the field of medical image analysis in recent years. CNNs and other sophisticated models like ResNet, DenseNet, EfficientNet, and Vision Transformers (ViTs) have proved to be more successful than traditional models in dermatological image classification, being more successful in terms of accuracy and efficiency. This article is a research-based discussion of the deep learning methods of psoriasis detection and a comprehensive discussion. We consider classic diagnostic issues, examine the AI models of the state-of-the-art and provide a deep learning architecture where CNN-based transfer learning and attention mechanisms are combined to enhance lesion classification. The experimental findings indicate classification rates over 96 percent, which implies clinical applicability of the DL-based dermatological diagnostics. Major issues, such as the diversity of the data, calculation costs, and interpretability of the model are critically analyzed. Lastly, we suggest some future research doctrines, such as explainable AI, expanded dataset, and integrating with precision medicine. This article highlights the importance of deep learning as the foundation of a new generation of dermatology, which will allow to diagnose psoriasis accurately, quickly, and at scale.