More averages improve MRI SNR with a square-root return, while taking proportionally more acquisition time. Larger acquired voxels contain more signal, but sacrifice spatial detail. Wider receiver bandwidth reduces SNR in the simple model, even though it can help with other imaging constraints. There is no single “best SNR” setting independent of the imaging task. [1][2]
What changes—and what stays fixed?
Signal-to-noise ratio describes the strength of a signal relative to random noise. It is one part of image quality. Contrast, spatial detail, artifacts, and acquisition time still need separate attention. A smoother-looking image does not automatically answer the clinical question better. [3]
The comparisons below hold the field strength, coil, sequence, TR, TE, flip angle, in-plane acquisition matrix, field of view, slice count, and reconstruction fixed. They illustrate conventional, unaccelerated 2D imaging of the same tissue with unchanged signal behavior and independent random noise. They are not a general model for 3D acquisitions or every scanner implementation.
| Change from baseline | Modeled SNR | The cost or qualification |
|---|---|---|
| Double the averages | About 1.41× | 2× acquisition time with the other timing factors fixed. |
| Quadruple the averages | 2× | 4× acquisition time. The improvement is not linear. |
| Halve acquired slice thickness | 0.50× | Finer through-plane sampling. Keeping the same number of slices changes coverage. |
| Double receiver bandwidth | About 0.71× | Shorter readout, but no automatic reduction in total scan time if TR and the sampling grid stay fixed. |
| Change the acquisition matrix | No single multiplier here | Voxel dimensions and acquired sample count change together. This is outside the learning tool’s fixed-matrix model. |
Ratios are relative to the same baseline, not measured patient SNR values. The square-root and voxel relationships are explained in the Stanford and university teaching references. [1][4]
Change a factor. Explain the result.
Predict the result before choosing a scenario. All inputs are multipliers of a shared baseline; they are not scanner settings to copy into a protocol.
Signal, detail & time
Fixed matrix and in-plane FOV. No acceleration, patient information, or data storage.
Illustrative 2D model. Results are theoretical ratios, not guaranteed scanner outcomes. The time factor covers averaging only; bandwidth and slice changes may trigger other timing changes on a real scanner.
The interactive controls need JavaScript. The comparison table and worked examples remain available without it.
A model, not a measured image value.
Other timing factors held fixed.
Choose a supplied scenario or reset the controls. No valid estimate is available.
Read the model and its assumptions
Time ratio from averaging alone = averages ratio
Thickness represents voxel volume only because the acquired in-plane voxel dimensions are fixed. The same tissue signal, coil, field strength, sequence contrast, sampling grid, slice count, TR, TE, flip angle, and echo-train setup are assumed. Signal averaging assumes independent random noise and unchanged signal between acquisitions. Changing those factors invalidates this particular comparison.
The time output isolates the contribution of averaging. It excludes setup, localizers, preparation, breath-hold overhead, sequence-specific timing changes, and additional slices needed to preserve coverage. It does not estimate the duration of a patient examination. Scanner software may adjust minimum TE, TR, slice packages, or other linked parameters when you change bandwidth or slice thickness; those effects are not simulated.
The formulas follow the fixed-parameter relationships in Stanford’s teaching material. Do not use these results to set patient-specific parameters or assess diagnostic adequacy.
Three examples worth being able to explain.
Why doubling NEX does not double SNR
Independent averages reduce the relative effect of random noise. The square-root relationship means that moving from one average to two gives approximately 1.41 times the SNR. Moving from one to four gives twice the SNR, with four times the acquisition time under the stated assumptions.
√2 ≈ 1.41 · √4 = 2
Averages may be labeled NSA, NEX, or a vendor-specific equivalent. Read the meaning of the control, not just its label. Averaging is not a general correction for motion or coherent artifacts.
Half the slice thickness. What restores the signal?
For an illustrative change from 3 mm to 1.5 mm, keep the acquired in-plane dimensions and other factors fixed. The voxel volume halves, so the model predicts half the original SNR. Four times the averages would restore that modeled SNR because 0.5 × √4 = 1.
0.5 × √4 = 1.00× SNR · 4× time from averaging
The through-plane sampling is finer, but this is not a free substitution. Slice coverage, cross-talk, and real sequence timing need separate evaluation. A model that holds slice count fixed does not preserve coverage automatically. [1]
Wider bandwidth can be useful despite lower SNR
Doubling receiver bandwidth at a fixed acquired matrix gives a modeled SNR factor of 1/√2, or about 0.71. A wider bandwidth can reduce chemical-shift misregistration and allow a shorter minimum TE. Whether it shortens the complete sequence depends on the rest of the timing—not the bandwidth value alone. [2]
1 / √2 ≈ 0.71× SNR
Do not compare an overall bandwidth in kHz directly with a per-pixel value in Hz/pixel. Even a ratio needs the same bandwidth convention in the baseline and modified acquisition.
When the simple SNR rules stop being enough.
A matrix change alters more than voxel size.
Acquired voxel volume is (readout FOV ÷ readout matrix) × (phase FOV ÷ phase matrix) × slice thickness. Increasing a matrix at fixed FOV reduces voxel dimensions, but also changes the sampling problem. Extra phase encodes can add acquisition time. This is why the lab does not offer a “matrix multiplier.” Interpolating an image afterward does not create the same acquired detail as collecting finer spatial information. [4]
TR, TE, and contrast are part of the explanation.
Changing TR, TE, or flip angle can change the signal and tissue contrast together. The calculator freezes those effects. If those concepts need a refresh, start with T1 vs T2 MRI and the role of TR and TE.
Acceleration and reconstruction need their own model.
Parallel imaging and partial Fourier change the sampling assumptions; compressed sensing and deep-learning reconstruction are also outside this model. Coil combinations and reconstruction filters can make simple image-background SNR measurements misleading. [6] This learning tool predicts relative behavior under limited assumptions; it does not measure scanner SNR or perform quality-control testing.
Do not optimize SNR in isolation. Unexpected noise, lost detail, motion, or an artifact may have different causes. Use the MRI artifact recognition guide to organize the questions, then follow the approved local protocol, device-specific conditions, and responsible MR team’s instructions. FDA guidance emphasizes site safety procedures and the conditions attached to MR Conditional devices. [5]
A reference for the next trade-off.
If these relationships are what you need to revisit, the MRI Scan Optimization Action Cards are the relevant MRITechStudio study resource: a problem-focused PDF reference on SNR, scan time, image quality, and artifacts.
Preview the cards on Etsy Read the product overviewThis free web lab is separate from the paid PDF. Check the Etsy listing for its current previews, included files, price, and terms. For a broader physics foundation, explore the MRI Core Concepts Workbook.
MRI SNR, NEX & bandwidth: quick answers.
Does doubling NEX double MRI SNR?
No. With the other factors fixed, the model predicts √2, or about 1.41 times the SNR, at twice the acquisition time. Four times the averages are needed for twice the SNR.
Does increasing receiver bandwidth increase SNR?
In the simple fixed-parameter model, it decreases SNR. Doubling receiver bandwidth gives about 0.71 times the SNR. The trade-off can still be useful for readout timing or chemical-shift behavior.
Does higher SNR always mean a better MRI?
No. A large voxel can improve SNR while hiding small spatial detail. Image contrast, artifacts, coverage, acquisition time, and the imaging task must also be considered.
Can I use this calculator to change a patient protocol?
No. It is a restricted educational model, not a scanner prescription, implant-clearance tool, QC measurement, or assessment of diagnostic adequacy.
Will more averages fix motion?
Not reliably. The square-root benefit describes independent averaging of random noise; it is not a general rule for removing structured motion artifacts. Review the cause and the approved workflow rather than treating every poor image as an SNR problem.
References & educational scope.
This guide uses established MRI physics rather than a vendor’s claim of clinical performance. The lab and worked scenarios are original educational implementations; they do not reproduce a manufacturer’s protocol or establish recommended settings.
- Stanford Lucas Center: SNR equations and parameter effects (PDF). Teaching slides dated February 12, 2015; fixed-parameter relationships and worked trade-offs.
- Siemens Healthineers Academy: MR Essentials—Image Quality. Manufacturer education on averages, slice thickness, receiver bandwidth, and scan-time factors.
- Perry Sprawls: Magnetic Resonance Imaging—Principles, Methods, and Techniques (PDF hosted by AAPM). Image detail and noise, especially printed pages 103–111.
- Larson Lab: Principles of MRI—Scan Time and Signal to Noise Ratio. University research-group teaching material; sampling time, voxel dimensions, and model limitations.
- FDA: MRI Information for Professionals. Site procedures, safety responsibilities, and device-specific MR conditions.
- Dietrich et al.: SNR measurement, multichannel coils, and reconstruction filters (2007). Original phantom and in-vivo measurement study in the Journal of Magnetic Resonance Imaging; DOI 10.1002/jmri.20969.
Source check: October 29, 2026. Prepared by MRITechStudio as independent educational content; no external clinician or medical-physicist review is claimed. Found an error? Send a correction. Review our editorial standards and educational-use terms.