Skin Diseases Detection Based on Deep Learning Models

Abstract

The diagnosis of human skin illnesses is the most ambiguous and challenging discipline of science. It has been noted that most cases go unreported due to a lack of improved medical infrastructure and equipment and the difficulty of visualizing the disease. This is where the concept for the project to create the first Android mobile application of its kind came from, which would assist any doctor or patient in performing a basic skin examination to identify diseases other than just skin cancer. With our work, we hope to reduce time, money, and effort involved in online diagnosis, which has become increasingly popular, particularly after the COVID-19 pandemic. Deep Learning (DL) is used in this study to build an efficient model using the DenseNet technique, which proved to be the most effective. The experimental results revealed that 87% accuracy, 83% precision, 83% recall, and 83% F1-score, which were used to evaluate the system performance, are the best outcomes that could be produced with the data collected. The software enables users to log in, drop the appropriate image for instantaneous detection from the gallery or their camera roll, and display the differential diagnosis of the lesion with percentages of each potential disease. Moreover, our proposed mobile application is introduced to cover the dermatologists in six different governorates around Egypt.