CRT students Pierre Murchan (cohort 2) and Shane O’Connell (cohort 1) have recently collaborated to write a review paper entitled “Deep Learning of Histopathological Features for the Prediction of Tumour Molecular Genetics”. In it, they outline some of the state of the art deep learning techniques used by researchers to predict the presence of biomarkers in digitised histopathological slides.
The goal of such research is for the techniques to be used in clinical settings to triage patients, saving valuable money and time. Many of these studies have achieved very high accuracy, however, others, while significant, require more development before they can be deployed to a clinical setting.
Researchers should be wary when reading articles which use deep learning techniques as any small changes in sample preparation or model fitting etc. tend to have huge impacts on outcomes. This can be seen in the differing accuracies in methods when applied to populations of different ethnicities from those that the networks were originally trained on.
The review directly relates to the project of Murchan whose research involves integrating genomics and histopathology data for subtype and treatment response prediction in upper gastrointestinal cancer patients. Also critical in this paper was O’Connell’s expertise in deep learning, gleaned from his PhD research.
These collaborations between CRT students emphasise the synergetic nature of the program where every student stands to benefit from the knowledge of every other student, even those belonging to different CRT cohorts. We expect to see similar cooperation between our other students in the future.