STIR listmode reconstruction : artifact when increasing subsets

[claudiadom]

Dear all,
I'm working with STIR to reconstruct data from a clinical scanner, and I'm trying two reconstruction approaches: from listmode and from sinogram data.
The problem I'm running into is with the listmode reconstruction: when the number of subsets is set to 1, everything works perfectly, but when I increase the number of subsets (keeping in mind that it has to be a divisor of the number of views), I start seeing an effect that looks similar to normalization not being applied correctly. I've attached two images showing the problem.

First image: 1 subset. Second image: 3 subsets. Third image: 11 subsets. My number of views is 297. These reconstructions used a low number of iterations.
Obviously, this effect doesn't occur when reconstructing from sinogram data, either with a single subset or with multiple subsets.
If anyone has run into something similar, or has an idea of what might be causing this, I'd really appreciate any help.
Thank you very much for your time.

Best regards,
Claudia

[KrisThielemans]

Dear Claudia

this looks like a bug, but can you provide some more information. A .par file would be good.

In addition, to help debugging this, could you try to set

  use_subset_sensitivities := 1
  ; for recomputing sensitivity, even if a filename is specified
  recompute sensitivity:= 1
  ; pattern for filename for reading the subset sensitivities, or writing if recomputed
  subset sensitivity filenames:=LM_subsens_{}.hv

and do the same (with difference filenames) for the sino-recon. The LM and sino ones should be identical.

[claudiadom]

Dear Kris,

Thanks for responding so quickly.

I've attached the two .par files corresponding to the listmode reconstruction and the sinogram reconstruction I was using.

I added this parameter, and now the resulting image for the listmode reconstruction looks good, without the artifact. It looks like this parameter makes the difference. But it seems strange to me that in the case of the sinogram reconstruction, the resulting image looked fine even without using use_subset_sensitivities.

These are two images reconstructed from listmode, and the only difference between them is this parameter. In the right image, use_subset_sensitivities := 1.

OSMAPOSL_emml_with_projmatrix.txt
OSMAPOSL_listmode_emml_with_projmatrix_bed1.txt

I can share the sensitivity images if you'd like

[KrisThielemans]

Great. I guess the default for that parameter might be different. It's definitely 1 for the sinogram case, but seems 0 for the list-mode case. That'd be a good (very small) PR for you to tackle :-). (Look for the set_defaults() function)

[claudiadom]

Dear Kris,

Thanks! I checked set_defaults() and both listmode and sinogram set use_subset_sensitivities = true by default, so I couldn't find where the difference would come from in the code; maybe it's something that happens elsewhere, or something I'm missing?

I also noticed that with 1 subset and use_subset_sensitivities := 1 or 0, the result is an image without artifacts. However, the artifact appears with listmode when using more than 1 subset and the parameter is set to 0. In the case of the sinogram reconstruction I never obtain this artifact.