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


The current study identified dermoscopic and confocal predictors for the diagnosis of in situ melanoma. Dermoscopic features independently associated with in situ melanoma diagnosis were an atypical pigment network and regression, with a risk three and four times higher of malignancy, respectively. An atypical reticular pattern and regression are already well–known dermoscopic diagnostic features of in situ melanoma;[6–10] However, the high frequency of these two dermoscopic descriptors even among atypical nevi suggests that with dermoscopy alone it is rather difficult to discriminate between atypical nevi and early melanomas with high sensitivity.

RCM is a valid method to assist in reducing the grey area between diagnosing nevi and in situ melanomas. Upon RCM, the finding of epidermal pagetoid spread, regardless of cell shape and location, confers a 2·8 times higher risk of melanoma; in the presence of junctional cytological atypia this risk increases to 3·4 (if focal) or to 8·4 (if widespread) times. In contrast, confocal findings of dense nests and melanophages are not associated with in situ melanoma diagnosis.

In the current study, the multistep diagnostic algorithm is able to discriminate between in situ melanomas and nevi with high sensitivity and specificity, 92·5% and 61%, respectively. The 94·9% sensitivity reached for the panel of early invasive melanomas, also guarantees the safety of the algorithm for deeper tumours.

Several diagnostic algorithms using RCM criteria have been proposed in past years.

Pellacani et al.[12] proposed a diagnostic semiquantitative algorithm for RCM evaluation of equivocal melanocytic lesions, in which confocal features associated with melanoma included two major features and four minor features. However, the study population included only 37 individuals with melanomas, of which four had in situ melanoma. Furthermore, two criteria were related to dermal features, namely cerebriform nests and cells within dermal papillae, and thus they are not present in 'in situ' lesions by definition.

In 2009, a two–step algorithm using RCM diagnosis of melanoma was developed by Segura et al.[16] In this algorithm, four features differentiated melanomas from nevi: two features of the dermoepidermal junction were protective and two features of the epidermis and papillary dermis were risk factors. Also in that study, invasive melanomas were included and thus the relative weight of distinct confocal features was different. Both the Pellacani et al. and Segura et al. algorithms have been tested retrospectively on a wider range of melanomas with different Breslow's thickness. If applied purely on thin equivocal lesions, the diagnostic accuracy would be significantly lower because some criteria are not present or readily identifiable. Furthermore, unlike our new algorithm, dermoscopic aspects were not considered. Finally, a more recent two–step algorithm described by Pellacani et al.[32] aimed to distinguish dysplastic nevi from melanomas and nondysplastic nevi. In the first step, lesions bearing epidermal or junctional cytological atypia and architectural disorders were selected to exclude nondysplastic nevi. In the second step, distinction between benign dysplastic nevi and melanomas was undertaken: presence of widespread pagetoid infiltration, diffuse cytological atypia at the dermoepidermal junction and nonedged papillae extending over at least 10% of lesions suggested malignancy. Although the latter score and the new algorithm share some RCM criteria, the previous score has a lower specificity compared with the new ones (data not shown) when evaluating thin lesions.

Overall, a common feature of the previous studies was the low number of in situ melanomas considered in the study population, which might account for lower specificity when considering thin lesions compared with the higher diagnostic accuracy of our new proposed algorithm. Moreover, dermoscopic features were not evaluated as part of the abovementioned algorithms. In clinical practice, a 'global' algorithm evaluating both epidermal and dermal RCM criteria is safer in order not to miss melanomas and in the case of flat, incipient lesions, the new score is much more specific and reliable, also considering the high sensitivity obtained from the small case series of thin invasive melanomas. Basically, clinical aspect and palpability might guide the clinician whether to apply one or another score to reduce the number of false–negative tumours.

Segura et al.'s[16] and Pellacani et al.'s most recent scoring systems were tested by Stanganelli et al.[33] and Lovatto et al.[34] on a small series (12 and 13 respectively) of melanomas, including a few intraepidermal ones. These studies looked at a narrowly selected population of lesions that had changed during dermoscopic digital monitoring and the patients had multiple atypical moles, thus limiting the score to extremely challenging lesions. However, the application of our new scoring system in these lesions (changing over time) could be the 'perfect scenario', where this new fast and reproducible RCM algorithm could be applied.

In line with pre–existing evidence, pagetoid spreading in the epidermis and cytological atypia at the dermoepidermal junction are confocal–positive predictors for melanoma diagnosis. Remarkably, the presence of cytological atypia at the dermoepidermal junction is more relevant than the architecture: this implies that a closer observation of single high–resolution RCM images should be carefully carried out so as not to miss subtle diagnostic details such as single atypical melanocytes.

Notably, when evaluating a suspicious lesion, the confocal finding of junctional cytological atypia, especially widespread, carried a higher weight compared with pagetoid spreading alone. In fact, the finding of pagetoid spreading when dendritic in shape, could be a false–positive finding as those cells cannot be identified with certainty as pagetoid cells but they could rather be Langerhans cells, a well–known confounder on RCM examination.[35] Whereas, the recognition of junctional atypical cells, roundish or dendritic–shaped, is undoubtedly related to atypical melanocytes and thus they represent a positive and reliable confocal marker of malignancy.

The strengths of the current study are the following: firstly, the dermoscopic and confocal criteria used in the algorithm are few and highly reproducible. Recent data[36] found that the presence of pagetoid cells and atypical cells at the dermoepidermal junction are the most reproducible RCM criteria among different evaluators. Secondly, the external validity of the algorithm was confirmed by the excellent diagnostic accuracy obtained from both the evaluation and the external validation sets, where similar sensitivity and specificity values were reached. Thirdly, including only five dermoscopic and confocal criteria that are organized in a multistep fashion (dermoscopic examination followed by confocal assessment, as routinely performed during a patient's care), ensures that our new algorithm will be easy to use in daily clinical practice. Fourthly, the number of in situ melanomas included in the current study is higher than that of the other previously published papers.

As with any algorithm, false–negative cases were found. In particular, nine out of 120 in situ melanomas were misdiagnosed and removed: two almost fully regressive lesions because of the confocal finding of junctional nests; two clinically atypical lesions because they were isolated, 'only sons of a widowed mother';[37] a dermoscopically atypical lesion with only focal junctional cytological atypia; a spitzoid lesion in a young woman and two slightly atypical lesions modified at dermoscopic follow–up and, finally, one nested melanoma in an elderly patient.

On the other hand, 83 out of 213 nevi were false–positives: a quarter of these were Spitz nevi and six cases were sclerosing nevi, both representing well–known confocal pitfalls; among the remaining 58 lesions, eight presented both an atypical pigment network and regression and almost all (79%) presented with pagetoid spread and/or junctional cytological atypia (90%) on RCM. Similarly to what happens with histopathological examination, in which a vetted plethora of criteria cannot always reliably distinguish in situ melanoma from other entities, with RCM analysis some nevi could be mistakenly classified as malignant.

Our study has several limitations. Firstly, the retrospective design is subject to recall and observational biases, which were addressed by involving three independent clinicians masked to the histopathological diagnosis. Secondly, the histopathological diagnosis of in situ melanoma might be highly problematic, as early melanoma might not display diagnostic criteria, and our cases were routinely assessed by two pathologists and not re–evaluated by a consensus pathological expert–meeting. Thirdly, we aimed to provide a combined dermoscopic and confocal algorithm, independently of clinical and epidemiological parameters.

To conclude, the multistep algorithm significantly improves the diagnostic accuracy of dermoscopy and RCM for the diagnosis of in situ melanoma and can be routinely used in tertiary referral centres equipped with RCM devices when dealing with thin lesions.