Predicting melanoma patient outcomes using digital pathology: Lucy Godson, 17/02/25

TIA Centre Seminar Series: Dr Lucy Godson

Full Title: Predicting melanoma patient outcomes using digital pathology

Abstract: Melanoma is the most aggressive form of skin cancer and fifth most common cancer in the UK. Identifying novel early-stage prognostic biomarkers and determining effective treatments are two key challenges for helping melanoma patients get better outcomes. Previous studies have analysed genetic data from tumours to stratify patients into immune subgroups, which were associated with differential melanoma specific survival and potential predictive biomarkers. However, this genetic analysis is not carried out in current clinical workflows, whereas haematoxylin and eosin (H&E) stained slides are routinely used in patient diagnosis. This talk will present our work on how deep learning models can be used to classify whole slide images (WSIs), into these molecular immune subgroups. I will discuss the application of different multiple instance learning (MIL) frameworks and examine how image resolution, feature extraction methods and aggregation strategies can affect model performance. I will also argue that graph representations can be used to encode spatial and contextual information within WSIs to improve immune subtype classifications. Finally, I will present our work on survival graph neural networks, for discovering new patient risk groups based on melanoma specific survival.