Abstract
DNA-based molecular assays for determining mutational status in melanomas are time-consuming and costly. As an alternative, we applied a deep convolutional neural network (CNN) to histopathology images of tumors from 257 melanoma patients and developed a fully automated model that first selects for tumor-rich areas (Area under the curve AUC=0.98), and second, predicts for the presence of mutated BRAF or NRAS. Network performance was enhanced on BRAF-mutated melanomas ≤1.0 mm (AUC=0.83) and on non-ulcerated NRAS-mutated melanomas (AUC=0.92). Applying our models to histological images of primary melanomas from The Cancer Genome Atlas database also demonstrated improved performances on thinner BRAF-mutated melanomas and non-ulcerated NRAS-mutated melanomas. We propose that deep learning-based analysis of histological images has the potential to become integrated into clinical decision making for the rapid detection of mutations of interest in melanoma.
