SKIN DISEASES PREDICTION USING CNN-VIT HYBRID MODEL

Abstract

Skin diseases pose a significant challenge to the health of the population that needs proper and prompt diagnosis to restrict its impact. This paper suggests a new hybrid model as a hybrid integration of CNN and ViT architecture in predicting skin diseases automatically. CNNs (local texture features extraction) and ViTs (global context is accounted in image features) have their strengths that we consider. The developed model with the cooperation of these complementary arch structures enhances the accuracy of diagnostic and generalisation ability of various skin diseases. Massive experiments were conducted using publicly accessible datasets of skin diseases, and the obtained results perform better than CNN or ViT models. The hybrid approach also exhibits high-resistance to noise and brightness of the input images and the colour and tone of the skin of patients which induces potential promise in practice. This paper illustrates the point of view of using CNNs and ViTs to continue working on the automated dermatology diagnostics that can contribute to improved and affordable healthcare among everyone.