Deep Learning Statistical Tests for Genetic Studies of Image Phenotypes

Christoph Lippert, HPI - Digital Health - Machine Learning Campus Golm, Building 28, Room 0.10810:15 - 11:15

In my talk I will give an overview on our ongoing work on the development of methods for finding associations between variation that is encoded in images and another variable such as genetics. To do so, we have developed statistical tests that leverage deep neural networks to represent the images and perform association testing on hidden-layer representations. This approach has enabled us to perform genetic studies of retinal fundus images in the UK Biobank cohort study.