All articles

Protocols

Bandwidth, Matrix and Averages: MRI SNR Trade-offs Explained

· By the Imaging Diversified Team · 7 min read

MRI

Every MRI protocol is a negotiation between signal, resolution and time, and three knobs usually move first: receiver bandwidth, matrix and averages. Understanding the MRI SNR trade-offs behind those three settings is what lets an imaging manager build protocols that hold up across scanners, shifts and staff changes. The relationships and figures in this article come from Siemens Healthineers' MAGNETOM Flash application articles, Allen Elster's Questions and Answers in MRI, Radiopaedia and Philips' Intera Release 10.4 Application Guide; the console examples apply to Philips systems, with generic names given so teams on other platforms can follow.

The rule underneath every MRI SNR trade-off

Joachim Graessner of Siemens Healthcare sums it up in a 2012 MAGNETOM Flash article on the SNR indicator: better resolution in less time with higher SNR is not possible without changing the field strength or the coil. Radiopaedia's trade-offs article makes the same point: SNR, resolution and acquisition time are linked, so changing one changes the others.

So every protocol change should answer one question: which of the three are we spending?

Receiver bandwidth: SNR against artifacts and timing

What bandwidth does

Receiver bandwidth sets how wide a range of frequencies the scanner listens to while it reads out the echo. The narrow bandwidth Q&A on mriquestions.com explains that a narrower bandwidth lets in less electronic noise, so SNR rises, and Graessner's 2012 MAGNETOM Flash article gives the relationship as SNR proportional to 1 divided by the square root of bandwidth.

The cost, according to the same mriquestions.com Q&A:

  • More chemical shift artifact, because each pixel covers fewer hertz.
  • Longer minimum TE, because each echo needs a longer readout, which can also limit minimum TR, echo spacing and slice count.
  • More distortion near metal and air–tissue interfaces.

Reading bandwidth on different vendors

This is where fleet protocols often go wrong. The receiver bandwidth Q&A on mriquestions.com notes that GE displays total bandwidth (for example 50 kHz), Siemens and Canon display bandwidth per pixel (for example about 195 Hz/pixel for 50 kHz over 256 pixels), and Philips displays water–fat shift in pixels. The same page gives the fat–water frequency difference as about 220 Hz at 1.5T and about 440 Hz at 3T, so a given Hz/pixel produces roughly twice the shift at 3T.

On Philips, according to the Intera Release 10.4 Application Guide (Volume 1, "Info page and Info line"), the Info page shows water–fat shift in pixels next to the matching bandwidth per pixel in Hz, along with the minimum water–fat shift the current sequence allows. The guide's parameter trade-off table says a larger water–fat shift (a lower bandwidth) raises SNR and artifacts without changing scan time or resolution.

Manager tip: when you standardize a sequence across vendors, record bandwidth in Hz/pixel and field strength on the protocol sheet, then convert it to each console's display. Copying "the same number" between a GE, a Siemens and a Philips scanner does not give the same bandwidth.

Matrix, FOV and voxel size: resolution against SNR

SNR follows voxel volume. Graessner's 2012 MAGNETOM Flash article notes that slice thickness changes SNR linearly. Its Table 1 shows that quartering the pixel area drops relative SNR to 0.25, before any change in phase-encoding steps is taken into account.

The Philips Intera R10.4 Application Guide puts numbers on the console-level changes:

  • Scan matrix: going from 256 to 512 cuts SNR to about 0.35 of the original (a 65% loss) and raises scan time linearly ("Scan matrix" section).
  • FOV: going from 250 to 325 mm raises SNR by about 69% and makes pixels about 30% coarser ("FOV" section).
  • Reconstruction matrix: interpolation only. It improves how the image looks, not the acquired resolution ("Reconstruction matrix" section).

This matters on any vendor. Philips' Info page lists both "measured" and "reconstructed" voxel size, according to the same guide. A protocol whose displayed pixels look fine may have been acquired much coarser. Build and compare protocols on acquired voxel size.

ChangeSNRScan timeWatch for
Lower bandwidth (larger water–fat shift)Up (1/√BW)No direct change; may lengthen if TE/TR must riseChemical shift, distortion, longer min TE
Larger scan matrixDown (256→512 ≈ 0.35×)Up, linearlyLower SNR at the same FOV
Larger FOVUp (250→325 mm ≈ +69%)No changeCoarser pixels
Thicker sliceUp, linearlyNo changePartial volume
More averagesUp (√NSA)Up, linearlyLong exams invite motion

Values in this table come from Graessner's 2012 MAGNETOM Flash SNR indicator article, the mriquestions.com bandwidth Q&As, Radiopaedia's trade-offs article and the Philips Intera R10.4 Application Guide, Volume 1.

Averages (NSA/NEX): the expensive way to buy SNR

Averages are the most predictable SNR lever and the most expensive one. Graessner's 2012 MAGNETOM Flash article shows SNR rising with the square root of the number of averages, so doubling them gives about 1.41 times the SNR while scan time doubles. The Philips Intera R10.4 Application Guide ("NSA" section) gives the same example: going from 2 to 4 NSA adds about 40% SNR for twice the time. The guide also notes that more NSA reduces motion and flow artifacts.

Before adding averages, check cheaper options such as a thicker slice or coarser phase resolution, and confirm the trade with the radiologist.

Partial Fourier and parallel imaging

Two time-savers sit on the same balance sheet:

  • Half scan / partial Fourier: the Philips Intera R10.4 Application Guide ("Halfscan" section) says it cuts SNR by about √2 and makes flow and motion artifacts more prominent. It also advises against combining it with NSA above 1, because reducing NSA is the better way to save time.
  • SENSE / parallel imaging: the same guide's trade-off table lists shorter scan time and lower SNR, with resolution unchanged. The higher the reduction factor, the bigger the SNR loss.

Use the console's SNR readout

Most consoles show a live relative SNR estimate. On Philips, the Intera R10.4 Application Guide ("Relative Signal Level") describes RSL as a percentage relative to the starting protocol, which begins at 100%. Halving slice thickness, for example, shows 50%. Graessner's 2012 MAGNETOM Flash article describes the equivalent SNR indicator on Siemens systems.

A checklist for any protocol change:

  • Write down the reason: speed, resolution or SNR.
  • Note relative SNR and scan time before and after the change.
  • Compare acquired voxel size, not reconstructed.
  • Record bandwidth in Hz/pixel and field strength for cross-vendor copies.
  • Check chemical shift and minimum TE after any bandwidth change.
  • Get radiologist sign-off and follow your facility's protocol change process.

For motion-specific fixes, see our MRI motion artifact checklist.

Key takeaways

  • Lower bandwidth raises SNR (1/√BW, per Graessner's 2012 MAGNETOM Flash article), at the cost of chemical shift, distortion and longer minimum TE.
  • Averages add SNR by √NSA while time rises linearly (Graessner 2012; Philips Intera R10.4 Application Guide).
  • Vendors display bandwidth differently (total kHz, Hz/pixel or water–fat shift in pixels, per mriquestions.com); standardize on Hz/pixel.
  • Build protocols on acquired voxel size and check the console's relative SNR readout after each change.

FAQ

Does lowering receiver bandwidth increase SNR?

Yes. According to mriquestions.com, narrower bandwidth admits less noise, and Graessner's 2012 MAGNETOM Flash article gives SNR as proportional to 1/√bandwidth. Expect more chemical shift and a longer minimum TE.

How much SNR do you gain by doubling averages?

About 41%. Graessner's 2012 MAGNETOM Flash article gives √2 (about 1.41) for doubled averages, and the Philips Intera R10.4 Application Guide gives about +40% for 2 to 4 NSA. Scan time doubles in both cases.

What is water–fat shift on Philips?

It is how Philips expresses receiver bandwidth: the fat–water displacement in pixels, according to mriquestions.com. A larger water–fat shift means a lower bandwidth, which the Philips Intera R10.4 Application Guide says raises SNR and artifacts.

Does a higher reconstruction matrix improve resolution?

No. The Philips Intera R10.4 Application Guide describes the reconstruction matrix as interpolation only, so acquired resolution is set by the scan matrix and FOV.

Building or harmonizing protocols across a mixed fleet? Reply with what you're working on, or request applications training and we'll work through it on your scanners. For more on chemical shift, see our guide to MRI fat suppression techniques.

Meet us at RSNA 2026

Our team will be at RSNA 2026 in Chicago (November 29 to December 3). If you'd like to talk about training your staff or better support for the technologists who scan, book a one-hour meeting with us, Monday, November 30 through Wednesday, December 2, between 10 am and 4 pm Central.

Book a meeting with us at RSNA

Sources

Get new imaging tips every week

One email a week with our newest articles for technologists. Free, unsubscribe anytime.

Share this articleX

Related articles