Melanoma is the deadliest type of skin cancer. Early detection is required to reduce mortality and improve health outcomes. A growing number of Artificial Intelligence (AI) models have been developed to aid in melanoma detection, and research shows promising results.
The aim of this retrospective study was to evaluate the value of including AI for melanoma risk prediction in the clinical assessment of skin lesions by a trained professional. The VECTRA WB360 whole body imaging system has two integrated AI models that assist with lesion malignancy risk evaluation (DEXI) and lesion tracking (Lesion Change Detection module). Both AI models are used for research purposes only. We applied both AI models to 46 melanocytic lesions in 16 high-risk melanoma patients who underwent VECTRA imaging as part of a randomized controlled trial conducted at the Princess Alexandra Hospital in Brisbane, Australia. The lesions were referred for excision and histopathological evaluation following clinical assessment, resulting in the diagnosis of 13 melanomas and 33 benign lesions.
Results show that 9 out of 13 (70%) melanomas were correctly identified by DEXI. The remaining four were flagged by Lesion Change Detection module, indicating potential malignancy. On the other hand, 10 out of 33 benign lesions were incorrectly flagged by DEXI. However, 7 of the 10 benign lesions were atypical/dysplastic naevi, which are known to exhibit characteristics suspicious for melanoma. Three of the 10 lesions incorrectly flagged by DEXI, were also flagged by Lesion Change Detection module, while six could not be assessed.
Our findings support the general notion that AI models have the potential to support or enhance human work instead of replacing it. Prior to clinical rollout, however, both models must be assessed and validated prospectively, and further research is required to define what levels of VECTRA assessed change constitutes clinically significant change.