JMIR Publications

ABSTRACT

Background:

Early detection of melanoma is lifesaving but remains challenging. Recent diagnostic studies revealed the superiority of artificial intelligence (AI) to classify dermoscopic images of melanoma and nevi, concluding that these algorithms should assist a dermatologist’s diagnoses.

Objective:

To investigate whether AI support improves the accuracy and overall diagnostic performance of dermatologists in the dichotomous image-based discrimination between melanoma and nevus.

Methods:

Twelve board-certified dermatologists were presented disjoint sets of 100 unique dermoscopic images of melanomas and nevi (total of 1200 unique images), followed by classification of images based on personal experience alone (part I) and with the support of a trained convolutional neural network (CNN; part II). Additionally, dermatologists were asked to rate their confidence in their final decision for each image.

Results:

While dermatologists’ specificity remained almost unchanged (70.7% vs. 72.5%; P=.54) with AI-support, sensitivity and accuracy increased significantly (59.4% vs. 74.5%; P=.003 and 65.0% vs. 73.5%; P=.002 respectively). Additionally, dermatologists’ confidence in their decisions increased when the CNN confirmed their decision and decreased when the CNN disagreed.

Conclusions:

This data indicates that CNN-based image classification is a promising and feasible tool to support clinicians in skin lesions classification and provides a rationale for studies of such classifiers in a real-life setting, where clinicians can integrate additional information such as patient age and medical history into their decisions.