MEGAPRESS Study Design Optimization

Dear Experts,

I am conducting pilot testing for an MRS experiment aimed at measuring GABA+ density in the visual cortex using the MEGAPRESS sequence on a Siemens 3T MRI scanner. Initially, we used the following parameters: TR = 3000 ms, TE = 68 ms, averages = 96, and a voxel size of 20x20x20 mm. This protocol was tested on 5 participants.

Upon analyzing the data with LCModel, we observed a substantial amount of noise, which prompted us to refine our protocol. In the next phase, we tested the following adjustments with one participant:

  1. We first increased the number of averages from 96 to 160, while keeping the voxel size at 20x20x20 mm.
  2. In the second iteration, we maintained 96 averages but increased the voxel size to 25x20x25 mm.
    Since we have other functional runs and a control region, we decided to keep the number of averages at 96 while increasing the voxel size. However, we are uncertain whether it would be wiser to sacrifice the control region and focus solely on the visual cortex. On one hand, many studies report results using only the visual cortex without a control region. If we decide to omit the control region, we could increase the averages from 96 to 160, and with the larger voxel size, we anticipate that the noise level would decrease further. On the other hand, we are unsure whether reporting the correlation between GABA+ density and behavioral performance exclusively in the visual cortex would be sufficient for publishing results.

I have attached the LCModel results for the different variations and our protocol https://drive.google.com/drive/folders/1jgB95C2n6cm3gx1rznzgX3wTBJnypVY-?usp=sharing.

I would greatly appreciate any comments or suggestions on how to enhance data quality and reduce noise in the most optimal way.

Thank you for your time and suggestions.

Hi Gizem and welcome to the forum,

I’ll come right out of the gate: Recommended for GABA-edited MEGA-PRESS is to start with a 27-ml volume. Even your larger voxel is just 12.5 ml (less than half), so you’d need to scan four times as long as the recommended 320 transients at TR = 2 sec just to get to the recommended SNR.

Yes, you’re getting a bit of signal back from going to TR = 3 sec, but you’re still way short of the recommendations. There is no replacement for SNR. You should increase your volume of interest or scan for longer. Being time-constrained by other planned acquisitions does not justify compromising on data quality (you’ll pay for short scans with decreased statistical power).

See also this thread for general Siemens protocol recommendations and these for MEGA-PRESS-specific ones on Siemens and Philips.

PS: Can’t access the Google drive, please change access rights.