203 Automated skin surface phenotype for melanoma risk assessment

While population screening for melanoma is not recommended, those at high risk could lead to earlier detection and improved patient outcomes. Current stratification largely based on subjective clinical assessment self-report. This time-consuming dependent the experience understanding of person reporting. Therefore, improved, standardized methods are required. We developed an automated skin surface phenotype which assesses phenotypes associated with increased from three-dimensional (3D) total body imaging. focused four phenotypes: i) color, ii) naevus count their distribution, iii) photodamage, iv) freckling density. 3D imaging provides a high-resolution capture almost entire within minutes. Images Queensland adults both general (n=156) high-risk populations (n=300) were annotated phenotypes. Skin color was assessed via image processing techniques, while convolution neural networks (CNNs) used develop classification algorithms counts, photodamage Results promising, overall accuracies > 85% (k=0.56) CNNs, suggesting can be accurately objectively extracted images. Validated against traditional calculators, combined stratify people into appropriate prevention early protocols, as well monitor changes over time, providing indications progression along naevus- melanoma-genesis pathway. novel approach combining CNNs overcomes current limitations improve workflow prioritizing highest developing melanoma.