How Were The Algorithms Tested?

Our skin lesion classification algorithm won the international ISIC 2018 Skin Lesion Diagnosis Challenge by a large margin. Participants were provided with a held-out test set to predict the diagnosis of 1512 skin lesions from 7 possible disease classes. Our method scored and balanced accuracy across all 7 classes with 88.5% and is now ready to undergo clinical trials for further evaluation of its use in a clinical setting. When complete, this holds the potential to be utilized as an invaluable second opinion tool capable of screening a large number of skin lesions, allowing for the most concerning cases to receive priority attention. However, the algorithms utilized within the Visual Search tool are to be used in an educational capacity to show similar cases and help as a learning tool for identification and matching.