Design a System to Automatically Detect Common Skin Diseases Using Deep Learning and Web Application

Skin and subcutaneous conditions affect nearly 1.9 billion people at any given time and are the fourth leading cause of the burden of non-fatal disease worldwide. Accurate diagnosis of skin diseases is also a challenge for dermatologists because several skin diseases have similar morphological features, extensive areas of inflammation, the complexity of skin color, and disease color change in hairy areas. Furthermore, the shortage and unequal distribution of qualified dermatologists significantly influences the diagnostic outcome. Based on a deep learning platform, this study developed an automated system to classify four common inflammatory skin diseases with similar morphological and color features including eczema, psoriasis, actinic keratosis, and severe acne. The study used 5433 images for four diseases taken from the SD-260 and Dermnet datasets. The deep learning model developed in this study was a customized ResNet152 model. The deep learning model developed in this study gave better than 89% accuracy results. Finally, we built a simple web application using Flask-Ngrok that automatically classified skin diseases with the developed built-in model. This study shows that deep learning has great potential for the accurate classification of skin diseases and strongly develops teledermatology to aid in disease diagnosis. Besides, it also provides a simple way to build web apps on Google Colaboratory, which anyone can access for free to build and run AI models independent of computer hardware.