Deep Learning Radiogenomic

Aims and Objectives

Radiogenomics (also known as Imaging Genomics) is based on the idea that entities at different scales such as molecules, cells and tissues are linked to each other and therefore can be modeled together; it aims to correlate genome wide molecular data with quantitative image features extracted from radiology images. Most of radiogenomics applications are in the field of oncology where radiogenomic biomedical data are routinely collected.

In this project, we propose a deep learning framework for radiogenomics analysis of cancer. The three main components of the framework are: imaging, genomic and integration. The imaging and genomic components extract relevant features from radiology images and genomic data respectively, while the integration component is responsible for finding correlations between the discovered sets of features. With this framework it is possible to incorporate multiple data sources, including human knowledge. Moreover, the components are deep learning models with a Bayesian formulation, making uncertainty quantification possible and model interpretability extractable.