In vivo Dermoscopic and Confocal Microscopy Multistep Algorithm to Detect In Situ Melanomas

S. Borsari; R. Pampena; E. Benati; C. Bombonato; A. Kyrgidis; E. Moscarella; A. Lallas; G. Argenziano; G. Pellacani; C. Longo


The British Journal of Dermatology. 2018;179(1):163-172. 

In This Article

Materials and Methods

Lesions with a histological diagnosis of in situ melanoma or nevus, excised with the suspicion of melanoma, were retrospectively retrieved from May 2011 to April 2016. The institutional review board of Reggio Emilia (protocol number 2011/27989) approved this study and all clinical investigations were conducted according to the Declaration of Helsinki principles.

Patient demographics and the location of lesions were recorded. Lesions located on the face and acral sites were excluded. Three dermatologists, blind for histopathological diagnosis, analysed all the images of lesions. Dermoscopic images were examined to assess the presence or absence of features suggestive of melanoma (atypical pigment network, blue–whitish veil, irregular pigmentation, irregular vessels, irregular dots/globules, streaks, regression).[25] Researchers were asked to score each RCM criteria reported in Table 1 as present or absent.[12–17,26–29] Also clinical features, such as palpability, colour (amelanotic, slightly pigmented and heavily pigmented) and size (more or less than 6 mm in diameter) were assessed for each lesion.

The study included a test set and a validation set. Test–set evaluations allowed the development of a multistep scoring system for the diagnosis of in situ melanoma. The validation set, matched according to age, sex and body site to the test set, aimed at assessing the performance of the multistep algorithm: lesions in the validation set were evaluated by three different dermatologists from those who undertook the assessment of the test set.

Furthermore, we tested the algorithm on a panel of 59 consecutive invasive melanomas (mean ± SD Breslow thickness 0·52 ± 0·2 mm), selecting all melanomas with a Breslow thickness < 1 mm diagnosed in our unit from May 2011 to December 2013.

Imaging Instruments

Dermoscopic images were captured by means of Dermlite Photo (3Gen LLC, San Juan Capistrano, CA, U.S.A.) equipped with a Canon G16 camera (Canon Inc., Tokyo, Japan). Confocal imaging was performed with a near–infrared reflectance–mode confocal laser scanning microscope (Vivascope 1500; MAVIG GmbH, Munich, Germany).[30,31]


Absolute and relative frequencies for clinical characteristics, dermoscopic and confocal criteria were obtained. Sensitivity and specificity got diagnosing melanoma were calculated. To analyse the dermoscopic and RCM factors influencing diagnostic accuracy, we used Spearman's rho coefficient to flag significant correlations, which were subsequently quantified via univariate logistic regression. A logistic multivariate regression model with backward stepwise variable selection was constructed to identify major independent factors among the descriptors that showed a significant difference (P < 0·10) on univariate analysis, together with the notable intervariable interactions. Alpha level was set at 0·05. Statistical analyses were performed using the IBM SPSS 22·0 package (IBM, Armonk, NY, U.S.A.). We created a pseudovariable by recoding 'nest type' in a dichotomous variable, which groups were 'dense nests' and 'others'.

The best cut–off point of the score was selected according to the accuracy, and receiver operating characteristic (ROC) curve and area under the curve (AUC) were calculated. The performance of the multistep score was tested on a validation dataset of 100 lesions. The ROC curve was generated and AUCs, sensitivity and specificity levels were calculated to evaluate the performance of the score on the external test set.