Recent advances in machine learning and image processing techniques have allowed to rapid annotation of medical imaging. The UKBB has a tremendous amount of imaging information acquired from diverse patients and already seminal work has been done to associate imaging traits with genetic features, however the annotation of imaging is the bottleneck. We aim to use deep learning methods (semantic segmentation models) to produce additional measurements and assessments of anatomic structures that could potentially be more precise than human measurements, and help inform further genetic studies to identify traits associated with disease and genetic influencers of disease. This project should take 3 years and produce algorithms and methods that can be generalized to conventional clinical imaging as well as help study the relationship between anatomy and genetics.
