Skin Disease Classification using Convolutional Neural Network via Android Smartphone Application

Abstract

According to the global perspective of health, skin disease leads a huge major burden. Due to a lack of medical awareness among the general populace, most patients did not notice the symptoms until several months later, allowing the disease to spread. Early detection of the skin lesions is important for making treatment decisions to prevent the spread of skin disease. The main purpose of this project is to develop an Android-based mobile application using convolutional neural networks that allow the users for diagnosing the skin disease easily for defeated the limitations of the conventional method. The model was executed using the TensorFlow library, which was implemented in the mobile application. The project used the MobileNet V2 model, with 6318 photos gathered from public dermatology sources on the internet. Acne, eczema, and vitiligo were selected as the disorders for detection. The overall accuracy of the smartphone application was 91 %. The detection rate was 100 % accurate, with an inference time of 317ms. The MobileNet V2 model is more preferable for smartphone device due to excellent performance, small size and fast training.