Detecting heritable phenotypic traits in images

Speaker: Christoph Lippert, HPI Potsdam

At the HPI Digital Health & Machine Learning research group, we are developing methods for the statistical analysis of large biomedical data. In particular we are interested in to detect heritable phenotypic features in medical images, as these are abundantly available and contain rich phenotypic information. While large imaging data sets are available, either from routine diagnostics or in large data repositories such as the UK Biobank, the analysis of imaging data poses new challenges for statistical methods development. Supervised learning with deep neural networks promise to automate the analysis and have been shown to be particularly suited for the analysis of medical images, but require large amounts high quality labels in the form of segmentation masks or disease annotations. Thus, manual labor by biomedical experts is a bottleneck to the analysis of imaging data. In this talk, I will give an overview over some of our current efforts in scaling up expert labeling by designing smart interfaces enhanced by deep learning, and using unsupervised or weakly supervised learning approaches to reduce the dependency on label information when associating images to genetic information.