FIX: Correct a number of problems introduced by changes for MIALSRTK v2.0.2

[pdedumast]

This PR proposes a number of changes that should correct all the problems introduced by PR #70 already merged into master for version 2.0.2. In particular, it includes:

• BUG: Some nodes get stuck in Running state but never process/finish #108
• Fix nipype nodes stuckes in running state #109
• BUG: Workflow ends with Segmentation Fault #111
• BUG: Fail build of docker bidsapp #112

[sebastientourbier]

@pdedumast I was indeed about to get to you about that! Doing so would act more as a refinement process, as for each new iteration, we initialize the transforms and update the reference image from the ones obtained from the previous iteration. But yes for sure, let's see how this works! 👌

[sebastientourbier]

@pdedumast The idea needs in fact to be better implemented as now the step length decreases after each iteration AND each image... which is wrong :/ So I got back to your old implmentation i.e. dividing by 4 the default values once and for all.

[pdedumast]

@sebastientourbier, based on the few tests I've made, reducing only once the registration step length seems enough to avoid slices that get completely lost. Though, further investigation might still be helpful to improve that.

Note: It seems that starting with big step length and then reducing it is not the correct strategy. I have observed strongly misregistered slices this way. That's why keeping a constant and not to large step lengths seems appropriate for now.

[sebastientourbier]

@pdedumast Just reviewed all the changes and it seems we are almost ready for the merge! But I noticed there is still a bug in the last execution of test 01 on circleci. When the pipeline is executed with the manual masks, a list of images defined by "stacks" in the param file, and the stackordering, all the images are then considered, while one would expect to take only the list of images defined by "stacks" in the param file.

[sebastientourbier]

@pdedumast I should have fixed the issue with the last commit cb7b769.

I added a preprocess.FilteringByRunid interface right after the datagrabber for the masks:

mialsuperresolutiontoolkit/pymialsrtk/pipelines/anatomical/srr.py

Lines 298 to 301 in cb7b769

	if self.m_stacks is not None:
	custom_masks_filter = Node(interface=preprocess.FilteringByRunid(),
	name='custom_masks_filter')
	custom_masks_filter.inputs.stacks_id = self.m_stacks

and review the inputs/outputs connection of nodes:

mialsuperresolutiontoolkit/pymialsrtk/pipelines/anatomical/srr.py

Lines 462 to 467 in cb7b769

	if self.use_manual_masks:
	if self.m_stacks is not None:
	self.wf.connect(dg, "masks", custom_masks_filter, "input_files")
	self.wf.connect(custom_masks_filter, "output_files", brainMask, "out_file")
	else:
	self.wf.connect(dg, "masks", brainMask, "out_file")

[sebastientourbier]

All tests passed on circleci for the lastest commit and after reviewing the changes introduced, this PR looks ready to be merged! Just wait from you to give me the go-head 👍
@hamzake: Could you test the singularity image built by circleci based on the latest commit (fea1a63): https://2023-210386291-gh.circle-artifacts.com/0/tmp/cache/mialsuperresolutiontoolkit.simg ?
@pdedumast: Have you noticed any issue that would need to be addressed in this PR before merging?

[pdedumast]

@sebastientourbier, looks good to me ! 👌