Principal Investigator: Professor Herve Lombaert
ETS Montreal, CanadaTags: 40254, active learning, geometry, Learning, semi-supervised learning, shape
The collection of medical imaging data is growing rapidly, faster than what humans can now process or analyze. In this context, geometry becomes essential to organize and exploit complex structures in medical images. Geometrical patterns, for instance, serve as strong predicators for critical diseases, notably in the brain and heart. The diversity and quantity of data in the UK Biobank provides a timely opportunity to study such geometrical patterns over large populations.
Aims & Objectives — This project specifically aims to advance new geometrical modeling of biological data in large databases, with three objectives over the next 3 years: (1) Build compact geometrical models of biological data, (2) minimize human effort while annotating images, and (3) further explore possible geometrical relations between imaging data and individual’s metadata, such as cardiac motion against behavioral information.
Method — We plan on developing computer-tools to highlight abnormalities among existing subjects in the UK Biobank. Deviations from common patterns in biology, indeed, potentially indicate risk of diseases. Our geometrical tools will directly assist the processing of large quantities of medical images. First, the geometrical statistics of brain, cardiac and abdominal scans will highlight common relations between imaging and behavioral data. This will find potential health conditions. Second, a recommendation system will exploit these relations to suggest which biological information should be further analyzed. This will improve human interpretation of medical images. Third, the cartography of imaging and individual’s metadata will reveal new conditions for health hazards. This will lead to a better understanding of how health and diseases vary across a population.
Expected Value — The research outcome is multifold. On methodology, the computational advantage will bring faster, more precise tools, notably in neuroscience and cardiology. On applications, recommendation systems will minimize human effort in annotating large-scale databases. On society, the developed tools directly act on automated predictions, thereby contributing to a better screening, diagnosis, and monitoring of individuals.