A smart dermoscope design using artificial neural network

Abstract

Melanoma is certainly the deadliest skin cancer. Clinicians try to detect melanoma at early stages in order to increase the successful treatment rate by using dermoscopes. We have designed a digital dermoscope that is both mobile and highly sensitive for automatic classification. We developed an accurate image processing software and a learning program that uses artificial neural network learning algorithm. A dataset of 200 images were used for training and 12 features were extracted. We considered common nevus, atypical nevus and melanoma as our diagnostic results. By doing that, we acquire three sensitivity and specifity values for each of the outputs. For the common nevus detection, SE = 100%, SP = 98.3%, for the atypical nevus detection, SE = 95%, SP = 97.5%, for the melanoma detection, SE = 92.5%, SP = 98.75%.