Background: Specific phenotypic traits are known to be associated with an increased risk of developing melanoma. The latest 3-dimensional total body photography systems integrate machine learning tools that automatically describe patients’ skin lesions. The aim of the present study is to describe machine learning outcomes from 3-dimensional total body photography and assess their correlation with phenotypic characteristics.

Methods: Cross-sectional study including patients imaged with 3-dimensional total body photography for which phenotypic information was available.

Results: A total of 633 patients with a history of melanoma, including 328 (51.8%) with atypical mole syndrome, and 5 (0.8%) diagnosed with Xeroderma Pigmentosum, were assessed. Machine learning tools generated an automated description of the patients’ skin surface, including automated counts, mean diameter, anatomical location, and main color of automatically detected lesions and nevi. Automated and in clinic nevus counts exhibited moderate correlation. Automated total lesion counts were significantly different between patients with a history of melanoma, melanoma and atypical mole syndrome, and Xeroderma Pigmentosum (612.3 vs 870.9 vs 6,865.5 respectively, p < 0.001). Patients with atypical mole syndrome had a higher automated nevus count than those without (190.9 vs 71.3, p<0.001). Carriers of the ‘RR’ MC1R red hair genotype showed significantly lower automated nevus counts compared to other variants (p <0.001) (Fig 2). Limitations include retrospective nature, single-center, and lack of generalizability.

Conclusions: Machine learning applications in 3-dimensional total body photography can provide an automated description of patients’ skin characteristics, which correlate with phenotypic traits that are associated with an increased risk of developing melanoma. These tools could enhance identification of patients at higher melanoma risk.