DC-SMIL: a multiple instance learning solution via spherical separation for automated detection of displastyc nevi

Abstract

Among skin cancers, melanoma is the most aggressive and most lethal form. Despite these terrible premises, an excision treatment carried out thanks to an early diagnosis is almost always decisive, guaranteeing the patient's survival. The early detection of melanoma is hampered by the extreme similarity of melanoma with other skin lesions such as dysplastic nevi. The current research is aimed at defining software solutions that support the computerized diagnosis of lesions for the detection of melanoma. To date, the proposals, both in terms of algorithms and frameworks, have focused on the dichotomous distinction of melanoma from benign lesions. However, the current debate on Dysplastic Nevi Syndrome (DNS), makes issues relating to the nature of the lesions, central to subjects who present a large number of moles throughout the body. In fact, individuals with DNS have a greater chance of being attacked by melanoma. The classification task relating to the distinction of dysplastic nevi from common ones is totally unexplored. In this document, we consider the difficult task of applying multiple-instance learning (MIL) approaches to discriminate melanoma from dysplastic nevi and outline an even more complex challenge related to the classification of dysplastic nevi from common ones. In particular, we introduce the application of a MIL approach that uses spherical separation surfaces. Since the results seem promising, we conclude that a MIL technique could be the basis of more sophisticated tools useful for detecting skin lesions.