Skin Disease Classification Using Deep Learning

Skin diseases pose a significant health concern globally, with diverse manifestations and diagnostic challenges. Deep learning methods have shown impressive potential in medical image analysis. This project focuses on utilizing methods of deep learning techniques for the automated classification of skin diseases. Two popular convolutional neural network (CNN) architectures, InceptionV3 and VGG16, are trained and evaluated on a comprehensive dataset consisting of around 17,000 images across 13 classes, including the most prevalent skin diseases such as acne, urticaria, eczema, psoriasis and vascular disorders. The study incorporates image preprocessing techniques to enhance the quality and informativeness of input data. Additionally, we experiment with the dense layers of the models, exploring configurations that optimize classification accuracy. The study aims to compare the accuracies of these models and determine the most effective one for deployment in a web interface for skin disease diagnosis. The proposed model, achieving a training accuracy of 80.88% for InceptionV3 and 74.17% accuracy for VGG16, demonstrates its potential as an effective instrument for healthcare providers and individuals, potentially aiding in the timely diagnosis and management of various skin diseases.