To improve the accuracy of subject-level activation detections in noisy fMRI data, models to optimize voxel-wise smoothness levels for both isotropic Gaussian filter and spatially adaptive steerable filters are proposed. The smoothing step with currently optimized FWHM is incorporated into the optimization algorithm and solved efficiently using a sequential quadratic programming solver. Results from both simulated data and real episodic memory data indicate that a higher detection sensitivity for a fixed specificity can be achieved with the proposed method as compared to the widely used univariate general linear models with various levels of smoothness.
This abstract and the presentation materials are available to members only; a login is required.