Early Skin Cancer Detection Using Deep Convolutional Neural Networks on Mobile Smartphone

Abstract

Malignant melanoma is the most dangerous kind of skin cancer. It is mostly misidentified as benign lesion. The chance of surviving melanoma disease is high if detected early. In recent years, deep convolutional neural networks have attracted great attention owing to its outstanding performance in recognizing and classifying images. This research work performs a comparative analysis of three different convolutional neural networks (CNN) trained on skin cancerous and non-cancerous images, namely: a custom 3-layer CNN, VGG-16 CNN, and Google Inception V3. Google Inception V3 achieved the best result, with training and test accuracy of 90% and 81% respectively and a sensitivity of 84%. This work contribution is mainly in the development of an android application that uses Google Inception V3 model for early detection of skin cancer.