A generative model for brain-age predictions
Approved Research ID: 65657
Approval date: November 9th 2020
The goal of this research is to develop a generative model for predicting age of healthy subjects based on their T1-weighted brain scans, in particular from grey matter segmentation maps. The predictive method entails a forward model that uses the target variable (age) and other demographic variables to predict the grey matter image of a subject. In a subsequent step, the model needs to be "inverted", to obtain the prediction of age, based on the subject's grey matter segmentation map and some demographic variables.
T1-weighted scans of healthy subjects from the UK Biobank will be processed to obtain grey matter segmentation maps, and then used to train and test the predictive method. The prediction performances achieved by the proposed method will be compared with the ones of current state-of-the-art methods for brain age prediction. The age prediction task is considered as a way to test the proposed predictive method, which can be used later on for prediction tasks in a clinical setting, such as prediction of some clinical scores in Multiple Sclerosis. Furthermore, brain age prediction is of clinical interest itself: several studies have shown that the difference between real age and predicted brain age correlates with some measures of disability. Therefore this "brain age gap" can be used as a biomarker, to study healthy ageing and to characterise pathological deviations underlying several diseases. The expected duration of the project is 2 years.