Segmentation of Prognostic Tissue Structures in Cutaneous Melanoma Using Whole Slide Images

Abstract

Our work applies modern machine learning techniques to melanoma diagnostics. First, we curated a new dataset of 50 patient cases of cutaneous melanoma in whole slide images (WSIs). We applied gold standard annotations for three tissue types (tumour, epidermis, and dermis) which are important for the prognostic measurements known as Breslow thickness and Clark level. Then, we devised a novel multi-stride fully convolutional network (FCN) architecture that outperformed other networks trained and tested using the same data and evaluated on standard metrics. Three pathologists measured the Breslow thickness on the network's output. Their responses were diagnostically equivalent to the ground truth measurements, showing that it is possible to overcome the discriminative challenges of the skin and tumour anatomy for segmentation. Though more work is required to improve the network's performance on dermis segmentation, we have shown it is possible to achieve a level of accuracy required to manually perform the Breslow thickness measurement.