Rakic Marianne

Master Thesis
Supervisors: Dr. Adrian Dalca, Prof. John Guttag and Prof. Ender Konukoglu

Anatomical Predictions using Subject-Specific Medical Data

Brain anatomy is intricately connected with many aspects of a person’s health. Anatomical change can provide important insight into neurodevelopment or disease progression. Accurate future predictions of such changes can be useful in designing treatments or scientific analyses. In this thesis, we present a method that predicts how brain anatomy will change over time. Specifically, from a single brain MRI and phenotypical information about the subject, our method can predict the appearance of a future MRI. We model anatomical changes through a diffeomorphic deformation field, and we design a function, using convolutional neural networks, that uses an initial scan and clinical data to predict these changes. Given a deformation field, a baseline scan can be warped to give a prediction of the brain scan at a future time point. We demonstrate our method using the ADNI cohort and analyse how the performance is affected when different subject-specific information is used.