Classification of Psoriasis Using VisionTransformer: A Deep Learning-Based Approach

Psoriasis is a chronic inflammatory skin disease that exhibits complex visual patterns, including erythema, scaling, and plaque formations, making accurate diagnosis challenging, especially in early stages. To achieve reliable and automated psoriasis classification, this article proposes a deep learning–based image classification framework using a Vision Transformer (ViT). The proposed method leverages the global feature modeling capability of ViT to effectively capture long-range dependencies and fine-grained texture variations in skin lesion images. A pretrained ViT backbone is employed and fine-tuned for binary classification of psoriasis and normal skin images. To address data imbalance, a class-weighted loss function is incorporated, while extensive data augmentation is applied to improve model generalization. The framework includes dataset splitting into training, validation, and testing subsets, adaptive transfer learning with frozen transformer layers, and optimized training using the Adam optimizer. Comprehensive performance evaluation is conducted using accuracy, precision, recall, specificity, F1-score, and confusion matrix analysis. Experimental results demonstrate that the proposed ViT-based approach achieves high classification accuracy and robust generalization performance, outperforming conventional CNN-based methods in capturing complex dermatological features. The results indicate that Vision Transformer models provide an effective and reliable solution for automated psoriasis diagnosis and can serve as a valuable decision-support tool in computer-aided dermatology systems.

Index Terms— Psoriasis classification, Vision Transformer, deep learning, skin disease diagnosis, medical image analysis, transfer learning.