Deep learning approaches to imaging genomics for precision medicine

This project will investigate the use of deep learning applied to imaging genomics to support precision health.

Precision medicine aims to tailor treatment to the individual, rather than assuming everyone will respond like the average patient. The biggest drivers in precision medicine have been developments in genomics. For example, knowing the genomic make-up of a tumour, e.g., lung cancer, allows clinicians to use highly effective targeted treatments against the tumour. However, a biopsy tissue sample is required to sequence the tumour genome, which is invasive and involves some risk. In addition, rapid mutation means tumours are often genetically heterogeneous. This heterogeneity is difficult to capture in a small biopsy sample, which can mislead and result in ineffective treatment.

Imaging genomics (sometimes known as radiogenomics) uses features derived from non-invasive medical images to infer the spatial distribution of the tumour genotype(s). Traditional imaging features have included the shape, greylevel intensity statistics, and texture of the tumour.

Deep learning is an artificial intelligence neural network technique based on multiple layers of neurons. It has had a huge impact on medical image analysis, setting the state-of-the-art performance in many benchmarks and applications, and outperforming human observers in some situations. This project will determine where deep learning can best be applied in the imaging genomic pipeline, to help ensure that every patient gets the right treatment at the treatment.