Multi-slice acquisition with 3D phantom

[aTrotier]

What happened?

Hi,
I would like to create a simultaneous multi-slice (SMS) example for a course.

First, I created with pypulseq a standard Multi-slice GRE sequence with long TR=1000ms in order to not worry about the spoiling.

The images are ok but with signal / contrast variation along the slice dimension

With phantom2D i have the same issue :

If I try with the EPI_MS from the example repository
I think I have the same issue (hard to tell with only 3 slices, I think it happens when I go further away from isocenter)

In order to change the slice position I am using this line of code in pulseq :

    rf_1.freq_offset = gz_1.amplitude * slice_gap * (sl - round((NSlices - 1) / 2))

Any idea ?

Environment

OS x86_64-linux-gnu
Julia 1.10.3
KomaMRIPlots 0.8.3
KomaMRIFiles 0.8.2
KomaMRI 0.8.2
KomaMRICore 0.8.3
KomaMRIBase 0.8.5

[cncastillo]

Hi! Hard to say. What do the different images mean? 20 slices are acquired simultaneously? or are the slices acquired sequentially?

Interestingly, the problem is also observed for the EPI_MS example; at least when simulating independently, the slice profiles have the same magnitude. Maybe a larger delay must be included between acquisitions for the magnetization to recover.

I will try running the MS EPI example using the development branch in #458, which should more precisely simulate RF pulses to discard a simulation accuracy problem

[aTrotier]

The slice are acquired sequentially. I also tried to acquire each slice in different sequence en get the same results :

If needed you can find attach the sequence (.ipynb and .seq) and reco (.jl)
GRE_MS.zip

[aTrotier]

@cncastillo do you still observe this on the dev branch ?

I finally have some time to create the SMS example

[cncastillo]

Hi!, sorry for not looking at this earlier. Thanks for the patience, I was submitting my thesis and I was quite busy.

In the latest version Koma v0.9 I get this:

GRE_SMS_slX.seq
X = 0, 1, 2, 3, 4

Is this more similar to what you expected?

[aTrotier]

Not really we still see the signal variation along the slice direction. Le jeu. 31 oct. 2024, 23:25, Carlos Castillo Passi ***@***.***> a écrit :

…

Hi, in the latest version Koma v0.9 I get this: GRE_SMS_slX.seq X = 0, 1, 2, 3, 4 image.png (view on web) <https://github.com/user-attachments/assets/55e37837-5a4d-4fbe-be67-f577291d19ea> Is this more similar to what you expected? — Reply to this email directly, view it on GitHub <#474 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AC5P7O2NMNIDMTFL6B6LP23Z6KU33AVCNFSM6AAAAABNUKAZOKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINJQHEZTQNJTGI> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[cncastillo]

Upon further examination, it seems that the phantom could be the problem. By default, brain_phantom3D is a thin slab of 2cm (z from -1cm to 1cm). The slice thickness (3mm) and slice gaps (10mm) of the sequence make it so the excitation goes beyond the phantom (x-axis in cm):

By modifying the phantom to

obj = brain_phantom3D(;start_end=[120, 240]) # z from -3 cm to 3 cm

I get:

Which looks a lot more uniform (XZ view vol[64,:,:]):

[aTrotier]

Great ! Do you already mention that in the documentation ? Maybe I should add a literate example about 2d multi slice or 3d with a large fov ? Le ven. 1 nov. 2024, 15:07, Carlos Castillo Passi ***@***.***> a écrit :

…

Upon further examination, it seems that the phantom could be the problem. By default, phantom_3D is a thin slab of 2cm (z from -1cm to 1cm). The slice thickness (3mm) and slice gaps (10mm) of the sequence make it so the excitation goes beyond the phantom (x-axis in cm): image.png (view on web) <https://github.com/user-attachments/assets/1925e91c-c028-4f4b-9016-9618f2bf4960> By modifying the phantom to obj = brain_phantom3D(;start_end=[130, 230]) # z from -2.5 cm to 2.5 cm I get: image.png (view on web) <https://github.com/user-attachments/assets/b575583e-acaf-45d1-9900-652c6bf65c7f> — Reply to this email directly, view it on GitHub <#474 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AC5P7O42YMWLV6H3YWWEKUTZ6ODIXAVCNFSM6AAAAABNUKAZOKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINJRHEZTEMJTGA> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[cncastillo]

The docstring for brain_phantom3D does not mention it, so that needs to be better explained. But it is mentioned here: https://juliahealth.org/KomaMRI.jl/stable/tutorial/04-3DSliceSelective/

A section at the end of that tutorial could be helpful, but ...

... even I forgot that brain_phantom3D was not the whole head. So maybe we can remove the start_end thing and always do the complete head. If a smaller slab is needed, the user can do obj[abs.(obj.z) .< 3e-2] (z from -3cm to 3 cm), for example. What do you think?

[aTrotier]

Or you can let the start_end but the default will be the whole brain ? Le ven. 1 nov. 2024, 19:09, Carlos Castillo Passi ***@***.***> a écrit :

…

The docstring for brain_phantom3D does not mention it, so that needs to be better explained. But it is mentioned here: https://juliahealth.org/KomaMRI.jl/stable/tutorial/04-3DSliceSelective/ A section at the end of that tutorial could be helpful. Now ... even I forgot that brain_phantom3D was not the whole head. So maybe we can remove the start_end thing and always do the complete head. If a smaller slab is needed, the user can do obj[abs.(obj.z) .< 3e-2] (z from -3cm to 3 cm), for example. What do you think? — Reply to this email directly, view it on GitHub <#474 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AC5P7O4MG4OA2XUDMWFO23TZ6O7V5AVCNFSM6AAAAABNUKAZOKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINJSGM2DMNRUGM> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[cncastillo]

Yeah it sounds reasonable. I will do that now, and update the tutorial accordingly.

Off course, If you have any cool example I would be glad to put it in the tutorials as well.