An Artificial Intelligence Tool to Detect and Classify Skin Cancer

Artificial Intelligence (AI) is being used to facilitate and automate traditional processes in different fields, including healthcare. Specifically, AI can be used to improve medical imaging and the diagnosis of skin-related diseases. In this work, we implement an AI-based solution to improve the current skin cancer diagnostic methods, including screening, self-examination, and microspectroscopy. These methods are costly, time-consuming, and depend on the availability of a professional physician. We conduct a scoping review to identify existing works on the used AI tools to diagnose skin cancer and we review 28 relevant attempts. However, there are still some issues with these techniques, such as accuracy, reliability, usability, and privacy. To overcome these issues, we build a deep learning model using the fast.ai Python library. The model is trained and tested on the ISIC dataset, which is anonymized and contains 11,720 dermoscopic images from seven different types of skin lesions. The model is then transferred into an iOS application that can be installed on any device with Apple’s A12 Bionic processor. Our tool overcomes issues in existing solutions as patients can securely take or import pictures of skin lesions using their iOS devices anywhere to get accurate diagnoses instantly. The tool is also affordable, compact, easy to use, quick, and provides diagnoses with relatively high performance. It outperforms the accuracy of existing works by 5 - 10% with a score of 95%. Existing works rely on one or two performance metrics to evaluate their models. However, to capture different aspects and to further ensure the reliability of our model, we include ten statistical measures with a grand mean score of 90%.