Figure 4.

Best practice in assessing techniques performances within the dermatological field. The first three guidelines were proposed based on the QUADAS-2 tool requirements, while the last one was derived by the literature review. The lesions chosen for the investigation should belong to a study population that reflects the standard population. The outcomes of a technique should be compared to the histopathological analysis of the lesion itself, except for trivially benign lesions, (in this case, dermoscopy can be used as an alternative). Indeed, histopathology is the current reference gold standard in this field, even if with its own limitations. As described in literature (86) the failure rate of histopathological analysis depends on the type of biopsy involved. Thus, excisional biopsy is advised. This approach stems from common clinical practice, albeit it may introduce possible biases in the classification trustworthiness of this type of lesions. It is known that the use of different reference standard for different lesion types may hamper the final evaluation of the performances of each technique, as well as the comparison with dermoscopy itself. The proposed dataset splitting is one of the main splitting methods used in this field, however, there can be others suitable for the specific task under investigation. MM, malignant melanoma; TP, True Positive; TN, True Negative; FP, False Positive; FN, False Negative.