Multi-Class Skin Disease Classification Using Convolutional Neural Network

Abstract

Our skin protects our body’s critical organs from potential harm by acting as an outer layer. This vital body part is prone to several illnesses caused by dust, allergies, bacteria, viruses, and fungi. Millions of people suffer from skin illnesses, which can result in discrimination and societal shame. Effective therapy for skin disorders requires a prompt and accurate diagnosis. Although advanced technologies like lasers and photonics have been used to diagnose skin diseases, they are frequently expensive and unaffordable for areas with low resources. As a result, image-based techniques present a quick and affordable substitute. Prior research has explored image-based diagnostic approaches for skin diseases. This study utilizes Convolutional Neural Networks (CNN) to classify various skin diseases, including Acne and rosacea, Lupus, Psoriasis, Tinea and Ringworm, Bullous disease, Cellulitis and Impetigo, Atopic dermatitis, vasculitis, Herpes, and Candidiasis. The image data underwent standardization in size, conversion to grayscale, and intensity balancing. Augmentation techniques were also applied. A comparative analysis used established CNN architectures like MobileNetV2 and ResNet 50. This project proposes a CNN architecture called Sequential Model, which is more accurate and reliable.