Principles of MR image formation and reconstruction.

This article describes a number of concepts that provide insights into the process of MR imaging. The use of shaped, fixed-bandwidth RF pulses and magnetic field gradients is described to provide an understanding of the methods used for slice selection. Variations in the slice-excitation profile are shown as a function of the RF pulse shape used, the truncation method used, and the tip angle. It should be remembered that although the goal is to obtain uniform excitation across the slice, this goal is never achieved in practice, thus necessitating the use of slice gaps in some cases. Excitation, refocusing, and inversion pulses are described. Excitation pulses nutate the spins from the longitudinal axis into the transverse plane, where their magnetization can be detected. Refocusing pulses are used to flip the magnetization through 180 degrees once it is in the transverse plane, so that the influence of magnetic field inhomogeneities is eliminated. Inversion pulses are used to flip the magnetization from the +z to the -z direction in invesrsion-recovery sequences. Radiofrequency pulses can also be used to eliminate either fat or water protons from the images because of the small differences in resonant frequency between these two types of protons. Selective methods based on chemical shift and binomial methods are described. Once the desired magnetization has been tipped into the transverse plane by the slice-selection process, two imaging axes remain to be spatially encoded. One axis is easily encoded by the application of a second magnetic field gradient that establishes a one-to-one mapping between position and frequency during the time that the signal is converted from analog to digital sampling. This frequency-encoding gradient is used in combination with the Fourier transform to determine the location of the precessing magnetization. The second image axis is encoded by a process known as phase encoding. The collected data can be described as the 2D Fourier transform of the object being imaged. Thus, a concept known as k-space is used to describe the image data and its relationship to the imaging gradient waveforms. The article demonstrates how phase encoding selects the row of k-space from which the data will be recorded, and frequency encoding determines the column. The goal of any acquisition strategy is to map k-space completely, and two methods, spiral imaging and echo-planar imaging, are described to demonstrate that the data acquisition path need not be a straight line.