Despite the large body of existing neuroimaging-based studies on dementia, in particular mild cognitive impairment (MCI), modelling and predicting the early dynamics of dementia onset and development in healthy brains is somewhat overlooked in the literature. Typical longitudinal studies are challenged by their requirement of multiple acquisition timepoints and the absence of inter-subject matching between timepoints. To address this limitation, we propose a novel framework that predicts the developmental trajectory of a brain image from a single acquisition timepoint, while classifying the predicted trajectory as ‘healthy’ or ‘demented’.
To do so, we first rigidly align all training images, then extract ‘landmark patches’. Next, to predict the patch-wise trajectory, we propose two novel strategies. The first strategy learns in a supervised manner to select training atlas patches that best boost the classification accuracy of the target testing patch. The second strategy learns in an unsupervised manner to select the most similar training atlas patches to the target testing patch using multi-kernel patch manifold learning. Finally, we train a linear classifier for each predicted patch trajectory. To classify the target subject, we use majority voting to aggregate the labels assigned by our model to all landmark patches’ trajectories. Our image prediction model boosted the classification performance by 5% without further leveraging any enhancing methods such as feature selection.
Student: Can Gafuroglu
Supervisor: Islem Rekik