Seminars: Survey Statistics

Minimizing Structural Errors in Synthetic MR Image Estimation

Abstract:

Title: Minimizing Structural Errors in Synthetic MR Image Estimation

Abstract: Existing approaches to signal estimation in magnetic resonance

(MR) imaging often result in training and prediction errors which are structural in nature. These are possible indications that we can we are not properly accounting for the spatial structure of the data. We propose a method which seeks to minimize these errors through a penalty function which penalizes first-order correlations between errors.  Training error is considered as well as expected prediction error under a first-order Gaussian Markov Random Field (MRF) model. Estimation of the optimal smoothing parameter will be discussed including cases for which there is not enough data to perform traditional cross-validation.