|
Abstract:
|
We propose a new spatio-temporal regression model for image response and image predictors that are acquired in a longitudinal framework for multiple subjects. The modeling approach predicts the current response at some voxel by accounting for the recent history of the predictors as well as for all the neighboring voxels. Our methods are motivated by a diffusion tensor imaging study of multiple sclerosis (MS), where of interest is to predict the magnetic transfer ratio (MTR) from conventional modalities: T1 and fluid attenuated inversion recovery (FLAIR) images. Acquiring MTR is not standard neuro-radiology practice due to the increased time and cost involved in the process. Nevertheless, scientific research has found that MTR maps, by being sensitive to the degree of myelination, both quantifies the tissue damage (e.g., white matter) and correlates with patient's disability, thus can be better tools for the diagnosis of MS.
|
