Digital dermoscopy analysis for automated diagnosis of pigmented skin lesions

Abstract

Melanoma is increasing in frequency worldwide. In Italy the current incidence of this tumor is about 10–15 new cases per year per 100,000 population.1 Despite much research into the field of melanoma biology and treatment that has already been done, this tumor is still practically incurable when diagnosed in an advanced stage. Early diagnosis is therefore of utmost importance to reduce the mortality rate.2 Clinical recognition of melanoma is, however, not always easy. Conventional clinical criteria for early diagnosis of melanoma are summarized by the ABCDE formula: A stands for asymmetry, B for border irregularity, C for color variegation, D for diameter 5 mm, and E for evolution (morphologic changes of the lesion). Although ABCDE clinical criteria represent a simple and reliable guide for self-examination, they cannot always be considered an effective approach to reach high diagnostic accuracy for melanoma, since numerous benign, pigmented skin lesions (PSLs) may mimic melanoma by using this method. From a diagnostic viewpoint, accuracy has greatly been improved since dermatologists have started to use new techniques, such as epiluminescence light microscopy (ELM), which magnifies lesions and enable them to be examined down to the dermo-epidermal junction.3,4 By ELM, new morphologic features of melanoma can be visualized, facilitating early diagnosis in most cases.5,6 Many studies have provided specific epiluminescence diagnostic criteria, paving the way for a new semeiology of patterns, colors, pigmentation intensity, configuration, regularity, and margin and surface characteristics. Qualitative evaluation of the many morphologic characteristics of PSLs observable by ELM, however, is often extremely complex and subjective.7 With the aim of obviating the problem of qualitative interpretation of dermatoscopic features, methods and instruments based on mathematical analysis of PSLs have recently been developed (Table 1).8–12 These methods are based on computerized analysis of digital images that are obtained by ELM. An example is digital dermoscopy analysis (DDA), which gathers numerical data and enables PSL images to be described objectively.