Automatic identification of skin lesions using deep learning techniques

Abstract

Cutaneous diseases or Skin Diseases are one of the most common diseases affecting nearly 2 out of every 3 people. However, WHO and the world bank records show that 50% of the world lacks access to essential healthcare. This is more prevalent for skincare. This lack of accessible skincare and a highly prevalent misdiagnosis of skin diseases demands alternate approaches to achieve universal skincare coverage. Technology holds the potential to bridge this gap between patient requirements and quality healthcare. Historically the research on utilising technology to provide dermatology care has been limited to teledermatoscopy and decision mechanisms on images. This research proposes a method considering disease images and the symptoms experienced in the diagnostics. Different deep convolutional neural network architectures are evaluated to choose the best one for an image-based classifier, and a feed-forward neural network for a symptom-based classifier, the results of each combined to yield the outcome. The ensemble method classifies the disease from symptoms and image with an accuracy of 87.71%. This approach can potentially be used to provide quality accessible skincare over the world through web and mobile applications bringing us one step closer in achieving the United Nations good-health and well-being sustainability goals.