Regularized super-resolution of brain MRI

In recent years super-resolution (S-R) methods are starting to emerge in the field of medical imaging for the reconstruction of isotropic images with increased slice resolution. Use of the maximal likelihood S-R estimator is not advisable as the S-R reconstruction is an ill-posed problem. Regularizing the S-R algorithm using specific apriori knowledge may compensate for missing measurement information and improve the resolved result. In this work two novel regularization methods are proposed, utilizing domain-specific spatial and intensity constraints on brain MRI data. Experiments indicate that the proposed methods eliminate disadvantages of common regularization methods and outperform these methods with better edge definition and increased image quality.