Overview

The objective of this presentation is to review basic and advanced image reconstruction methods that are used clinically. In the prior lectures in this series, you have focused on how to generate, encode and acquire MRI data. In this lecture, we will focus on how to take the raw MRI data and generate images.

The data acquisition process in MRI involves applying a sequence of RF excitations and gradients; this process inherently encodes data in the spatial frequency domain (or k-space), where each data point in k-space contains information about the entire object. As a result, the raw data in MRI are not collected directly in the image domain, but instead must be transformed in a process known as image reconstruction. Most often, image reconstruction in MRI is performed by calculating the 2D inverse Fourier Transform of the k-space data.

In this presentation, we will first discuss the relationship between k-space and image domains, as well as standard 2D and 3D Fourier reconstructions, since these concepts are fundamental to understanding image formation in MRI. We will review important characteristics of k-space and different k-space sampling trajectories (both Cartesian and non-Cartesian), and we will discuss how these affect the appearance of MR images. We will then explore how these properties can be exploited in more advanced reconstruction methods that are commonly used clinically to reduce imaging time and improve image quality. While there are many reconstruction methods (and many more will be presented at the current meeting), here we will focus on a few major categories of reconstruction methods that are employed on clinical MRI scanners. Specifically, this section will include a short overview of partial Fourier techniques, parallel imaging, compressed sensing, and deep learning.

Acknowledgements

No acknowledgement found.

References

No reference found.