Understanding the genetics of human physiology by integrating imaging and genetic data in UK biobank
Approved Research ID: 65439
Approval date: November 10th 2020
The goal of this project is to perform automated measurement of skeletal traits from medical images using deep learning. This will allow us to combine this information with genetic data to understand the genetic basis of these traits.
Impact and rationale
Imaging data is standard of care in clinical practice, and have been collected routinely in millions of people. Developing methods to automatically convert the imaging data into quantitative measurements will then also allow us to examine the relation of these datasets to disease status.
Specifically, we will take X-ray images from different parts of the body (whole body, knee, hip, spine) and then build machine learning methods to segment each of these images to specific bones (at pixel level resolution). We will then measure these individual bones and perform a genome-wide association study to determine genetic variants that are associated with these measurements. Finally, we will combine genetic data, imaging phenotypes and disease status of osteoporosis or osteoarthritis available from the health record data. This will allow us to build predictive models for risk in these important musculoskeletal disorders, as well as examine causal relationships between the imaging and genetic data, which and final disease outcome.