Development of a Mobile Application for the Early Detection of Skin Cancer using Image Processing Algorithms, Transfer Learning, and AutoKeras

Abstract

Skin cancer is one of the most common and dangerous types of cancer. With global ozone levels depleting and more ultraviolet radiation reaching the Earth's surface, rates of skin cancer are predicted increase rapidly. As per WHO, around 3 million cases of skin cancer are diagnosed every year which lead to thousands of deaths. The most important step in skin cancer treatment is early and accurate diagnosis when the survival rate is high, and successful medical treatment is possible. But with current tools, the skin cancer diagnosis process is subject to errors and results to be inaccurate, inefficient, and not globally scalable for developing and underdeveloped countries. This research proposes, SkinScan, a novel mobile application that uses deep learning to efficiently and accurately diagnose the 7 main types of skin cancer. This application utilizes a fine-tuned EfficientNetB7 CNN model that was found to be the most optimal after a comparative analysis of ten different CNN architectures. This chosen model had the highest validation accuracy of 95% and F1 score of 0.94. SkinScan's supplemental features include self-assessment tests for skin cancer risk, protective guidelines for exposure to UV radiation, and thorough information about each of the types of skin cancers, and their symptoms and treatments. SkinScan is an all-in-one that can significantly mitigate skin-cancer rates around the world by providing early skin cancer diagnosis.