Abstract

Magnetic Resonance Imaging (MRI) is the method of choice for diagnosis, evaluation and follow-up of brain pathologies. In the common treatment scheme, patients are repeatedly scanned every few weeks or months to assess disease progression and treatment response. Although the important information for clinical evaluation lies in the change between the follow-up MRI and the former one, every follow-up scan is acquired anew. This makes most of the data in the later scan redundant.

In MRI , data is acquired in a spatial frequency domain, called “k-space”. In my talk I’ll discuss the application of compressed sensing (CS) for MRI and the mutual similarity of follow-up scans in longitudinal MRI studies. I’ll present a sampling and reconstruction framework that exploits the redundancy of the acquired data in longitudinal studies. This would rely on two extensions of compressed sensing, adaptive-CS and weighted-CS. In adaptive CS, k-space sampling locations are optimized such that the acquired data is focused on the change between the follow-up MRI and the former one. Weighted CS uses the locations of the nonzero coefficients in the sparse domains as a prior in the recovery process. Results are presented on MRI scans of patients with brain tumors, and demonstrate improved spatial resolution and accelerated acquisition for 2D and 3D brain imaging at 10-fold k-space undersampling.