In MRI, we sample the object’s frequency (rather than physical) domain. The number of possible sampling strategies is, in this case, equal to the number of ways one can traverse a three dimensional space – hence almost unlimited. Still, 99% of MRI scans employ a conventional Cartesian grid, due to its robustness to hardware inaccuracies, and due the simplicity of reconstructing images using fast Fourier transform as opposed complexity of reconstructing non-Cartesian data. This being said, non-Cartesian sampling can be highly advantageous owing to several unique characteristics:
 

• High robustness to motion artifacts, leading to sharper imaging point-spread function

• Immunity to aliasing (“ghosting”) artifacts, allowing to image at arbitrary field-of-views, and thereby offering higher spatial resolutions.

• Incoherent undersampling artifacts, which makes this sampling scheme an ideal for accelerated scans techniques such as compressed sensing or model-based reconstruction.