What is the proper value to the SNR/tSNR of multi-band BOLD image?

Hi everyone,

Our group is working on optimizing the EPI parameter for a better BOLD quality. We found that the signal-to-noise ratio (SNR) and temporal SNR (tSNR) decreased sharply when using multi-band method to accelerate the TR.

For example, the SNR/tSNR were 6.0/80 to the original BOLD image, but they decreased to 3.4/50 using 3 acc. factor. Therefore, I am wondering why it would happened and if there is a “reference value” for the multi-band image’s SNR/tSNR.

Any ideas would be appreciated !


It might help if you included the BIDS sidecar JSON files to provide a bit more details. In particular, you need to be cautious with low-dimensional coils and also make sure that the number of slices divided by the MB factor is an odd number, both described here. Even if all else is equal, for fMRI MB is a tradeoff of increased number of observations versus decrease SNR per sample. This is simply because the faster TR provides less time for T1-recovery. This article as well as this one provide more details on the considerations for using MB. For many questions with recent hardware, moderate MB factors are worthwhile, and for some designs it can help you avoid aliasing effects of physiological noise.


Thank you for your kind help!

The scanner and system we used is GE SIGNA Architect with a 48-channel coil. We set the phase acceleration to 2 while the hyperband slice to 3. The slice is 60, so the slices/MB factor = 60/(2*3) = 10, or 60/3 = 20. Please check the MRIQC output html for other parameters for BOLD image without multi-band (this) and with multi-band (this).

Thank you for explaining the reason for the SNR reduction in multi-band EPI sequence to me. I learned that some risks are worth taking in specific research goals, like dynamic analysis in fMRI which requires a dense sampling in time scale. And I appreciate you for pointing me the references, I will make them useful in my studies.


The BIDS JSON files do contain additional sequence information. Regardless, as noted by Barth et al. you will want an the slices divided by multi-band to be an odd number (e.g. 62 slices for MB=2 and 63 slices for MB=3). Given you are a GE system, @mr-jaemin may have some additional suggestions.

1 Like

Please always consider a total acceleration MB (3) x phase (2) = 6, not just MB acceleration (3) because both acceleration (MB and phase) is essentially based on GRAPPA (ARC).


Thank you for your prompt about the MB factor. I will keep that in mind.

Would you mind me asking a quick following question? From you experience, if I need the acceleration factor to be 4, which would you recommend between setting MB(2) x phase(2) or MB(4) x phase(1)?

Thanks in advance!

Oh… I am sorry for my carelessness. I re-upload the JSON file generated by dcm2niix. Please check the JSON file of the single-band BOLD image here and multi-band image here.

I appreciate for your suggestions. I will consider modifying the slice to meet Barth et al. recommendations. Besides, if I want to reduce the slice to MB(6) x 9 (odd number) = 54, the space gap would be introduced in the sequence (We did not use space gap for now, by 2.5 mm thickness with 60 slice). What do you think about using the space gap in SMS sequence? Would the space gap decline the effect size or reliability about the results?

Thanks inadvance : )

I personally prefer to use MB=2 x phase 2 on Architect. When you want to use a larger matrix (frequency) + fixed TE without phase acceleration on Architect, partial Fourier may be used. To confirm/avoid this, please double check the “Minimum TE” (shown in the bottom) with the “Min Full” selection, not with a fixed number (e.g. 30ms), making sure that your desired TE (e.g 30 ms) is larger than “Minimum TE” (e.g 25 ms).

Note with a fixed number, the “Minimum TE” displayed is with partial Fourier.
Note with the “Min Full”, the “Minimum TE” displayed is with Full Fourier.

From the JSON data collected, you will see “PFF” as a part of “ScanOptions” if partial Fourier was used. I don’t see this from your JSON files.

1 Like

Thank you very much! I will modify the parameters as suggestions.