Automated skin lesion screening--a new approach

Abstract

Automated melanoma diagnosis is a popular focus of research, with numerous papers describing techniques and results. In our study, we identified two possible problems with the current method of automated diagnosis, where systems are intended to reproduce histopathology results. We propose a new method of identifying problematic skin lesions, namely attempting to reproduce algorithmically the perceptions of dermatologists as to whether the lesion should be excised. In the best case, our initial model reproduced the decision of dermatologists in over 80% of cases. These results suggest that reproducing the decision to excise may be a valuable adjunct to current methodology.