Registration of nevi in successive skin images for early detection of melanoma

The only cure for malignant melanoma (skin cancer) is early detection. Surgical removal of a newly developed melanoma will result in complete cure. In this study, the first steps towards the development of a skin cancer detection system are reported. One possible way to detect early the occurrence of melanoma is to screen the body of a patient at regular intervals for changes or new lesions. To increase the accuracy of this laborious and painstaking task, a computer vision system could be used. One of the most important problems involved in a vision system of this kind is the need to determine which lesions in successive skin images taken over a given period, represent the same lesions. Repeated registrations of skin images also detect new lesions that do not have a counterpart on the previous image. After registration, the lesions in successive images are compared for alterations in size, shape, colour and so on, to detect changes that are suggestive of melanoma. In this paper, we introduce a new algorithm for the registration of lesions in successive skin images. The baseline algorithm requires two initial matches to register the other lesions in the images. The initial matches are provided by a physician or an algorithm that selects the most likely initial matches. The test suggests that the baseline algorithm determines 99% of the matches correctly, and this performance is largely independent of the number of lesions in the skin images.