Deep learning can see the unseeable: predicting molecular markers from MRI of brain gliomas.

This paper describes state-of-the-art methods for molecular biomarker prediction utilising magnetic resonance imaging. This review paper covers both classical machine learning approaches and deep learning approaches to identifying the predictive features and to perform the actual prediction. In particular, there have been substantial advances in recent years in predicting molecular markers for diffuse gliomas. There are few examples of molecular marker prediction for other brain tumours. Deep learning has contributed significantly to these advances, but suffers from challenges in identifying the features used to make predictions. Tools to better identify and understand those features represent an important area of active research.