Artificial Intelligence Image Recognition Inhealthcare

摘要

Skin melanoma is one of the most dangerous cancer tumor forms. The main reason is not only it(s aggressiveness but also uncontrolled growth. The death(s number due to this cancer tumor rapidly increased in last 20 years, doubled every 10–15 years and shows 7% annual growth. The MIPT(s Special-Purpose Digital Systems Laboratory proposes a new convolutional neural network based algorithm of skin deceases identification. This method provides to reach the classification accuracy of 94% at dermoscopic pictures of skin decease and about 88% at microscopic ones. The highlight of the method is limited training set working ability. Whereas overwhelming variety of neural network-based algorithms demands 10 000 and more pictures to train, algorithm proposed could be operated with a training set of 1000 pictures with a declared accuracy. In February 2018 the laboratory started free web service in testing mode available at https://skincheckup.online