Abstract:
|
We introduce a scalable, multi-resolution spatio-temporal model and a computationally efficient methodology to estimate task-related motor activation and whole-brain connectivity with the aim of monitoring motor function recovery of stroke patients. The proposed model allows to test for voxel-specific activation of Blood-Oxygen-Level-Dependent response while accounting for a non-stationary local spatial dependence within pre-specified Regions Of Interest (ROIs), as well as global (among ROIs) dependence. The model allows for flexible functional estimation of the region-specific haemodynamic response and is able to detect activation patterns among ROIs via graphical LASSO estimation of the inverse covariance matrix. Inference is performed on an fMRI data set with 150,000 voxels per time frame, for a total of 22 million data points, which a multi-step methodology explicitly designed to scale on high performance clusters.
|