Myths of basic EPI

Uncategorized 5 Minutes

Authors: Kenny (Chung) Kan1, Rüdiger Strinberg2, and Renzo (Laurentius) Huber3

  1. Functional MRI Facility (FMRIF), NIMH, NIH, USA
  2. German Center for Neurodegenerative Diseases (DZNE), Germany
  3. Martino’s Center for Biomiedical imaging, Mass general Brigham, Harvard Medical School

What happens with the EPI sequence when I increase the TE? Which bandwidth should I use? What do I need to change to reduce PNS? There are many common misconceptions in basic EPI protocols. This blog post discusses the most common ones.

This post is a a continuation of highlighting basic acquisition features of EPI. See posts in GRAPPA, Regularization, Ghost correction, and 3rd order shim here.

On a first glance, Echo Planar Imaging (EPI) is simple enough to be described with a rectangular back and forth pattern k-space (left of Fig. 1). But still, there are hundreds of protocol parameters that go into it.

Fig. 1. Paradox that a simple zig-zag space trajectory needs so many protocol parameters to be determined.

As a scanner operator, this is still not enough to have full control over the most critical components of EPI; we usually only have one parameter that automatically controls 7 independent features of a given EPI readout pulse. These readout parameters are additionally interrelated to the properties of the encoding blips that are played out between the EPI readouts (at least three more parameters).

While the details of how all of these parameters are controlled automatically may differ between different EPI implementations, the underlying physics and hardware limitations are the same. Here we focus on the SIEMENS product EPI sequence (ep2d).

Fig. 2 Paradox that the scanner operator does not have control over the most basic pulse shape parameters. Many aspects of the pulse shape are inherently estimated from one single protocol parameter: Bandwidth.

Myth or Truth? Test your knowledge of what the parameters do?

What happens when you increase the TE to be longer than the minimal TE?

Fig. 3: What happens with the sequence, when you increase the TE to be longer than the minimum TE? A) does it slow down the readout? B) does it include dead time?
Click on the figure to reveal the solution. (this will open the picture with the solution in a new tab that has additional text on the figure revealing the truth).

What happens when you relax the Gradient mode?

We often need to change the gradient mode, when we run into issues of peripheral nerve stimulations or when we cannot achieve the echo spacing we need.

Fig. 4: What happens to the sequence when the gradient mode is changed to “Normal”? A) is the gradient slower and weaker? B) Is the slew rate smaller, resulting hin high amplitudes to achieve the same gradient moment for a given bandwidth?
Click on the figure to reveal the solution (this will open the picture with the solution in a new tab that has additional text on the figure revealing the truth).

Where are the gaps in the Bandwidth spectrum come from?

Why can’t we adjust the bandwidth to any value that we like? There are many gaps and borders. Whats keeping us from using bandwidth values within these gaps? Why can we not use larger bandwidths?

Fig. 5: There are many reasons, why bandwidth are not selectable?
Click on the figure to reveal the solution. (this will open the picture with the solution in a new tab that has additional text on the figure revealing the truth).

Explanation: The Myth is that these gaps and borders are from “forbidden” frequencies. These are bandwidth that result in echo spacings that would result in mechanic (aka acoustic) resonances that could damage the gradient. While these “forbidden” frequencies are indeed causing some gaps in the spectrum of selectable Bandwidths, there are other reasons too.

To understand the other ones, we need to think about how the ADC (analog to digital) works. You can think of an ADC as a series of little capacitors. They are open for a short amount of time (dwell time) to store the voltage. After this time window has passed, the stored voltage is read out digitized as stored in memory. However this procedures has limitations:

  • The capacitor cannot be open for infinitesimally short amount of times. With a typical ADC bandwidth of 500kHz (1MHz with 2fold oversampling). This limits the shortest dwell time. the shortest time it can be open is 2μs. Dependent on the FOV uses, this can limit the maximal selectable bandwidth.
  • The dwell time cannot be controlled with an infinitesimally accurate clock. E.g. the ADC clock can allow dwell times of 2μs, 2.1μs, 2.2μs, 2.3μs, 2.4μs. But values like 2.15μs are not possible. This dwell time increment limit of 100ns is responsible of the series of little gaps that are typically visible at the higher end of the bandwidth.
  • The maximum capacity of the ADC is also limited. It can only store a limited amount of capacity (F). So there are limits in place that cap the maximal dwell time (0.1ms). This can be the lower limit of the Bandwidth.

Slew rate increases with increases bandwidth?

Increasing the readout pixel bandwidth at a given resolution means that the ADC gets shorter (BW ∼ 1/ADC duration). So, what happens to the gradient pulse when we increase the bandwidth? Is the slew rate increased to achieve faster EPI sampling?

Fig. 6: EPI gradient pulse shape as a function of bandwidth. It can be seen that the bandwidth does not affect the slew rate at all (aside of small rounding effects). With increasing bandwidth, the sequence utilizes both, a) stronger gradients and b) more ramp sampling.
As soon as the maximal gradient amplitude is reached, the bandwidth cannot be further increased at the used ramp sampling ratio. In these cases, the amount of ramp sampling gets less, i.e. the gradient flat top time has to increase to achieve the same area under the curve while having shorter ADC durations (e.g. see red and blue cases).

This can also be seen in the graph below showing the echo spacing as a function of bandwidth.

Fig. 7: Echo spacing and peripheral nerve stimulation as a function of bandwidth for FOV 160mm and 2mm resolution using a SIEMENS SC72 gradient system (VE12U). It can be seen that the echo spacing decreases with increasing bandwidths. The non-linear dependence arises from increasing ramp sampling ratios along the downwards slope. Note the blocked echo spacings due to “forbidden” frequencies. Small scattering of values comes from minor variations of slew rate to both implement ramp sampling and accommodate phase encoding blips (on gradient rater time = 10 μs) between EPI readouts at the same time.

Acknowledgements and Conflict of interest

This blog post is specific to SIEMENS EPI. Other MRI console providers might have their protocol parameters differently.

Kenny is supported by the NIMH/NIH (ZIC MH002884).

Unknown's avatar

Published by kanc

Published

Leave a comment