I am working on running resting state fMRI pre-processing on a cohort of human subjects.

My goal is to use either of two conventional SPM-based protocols:

1. EPI realignment, EPI-subjectT1 co-registration, 1-interpolation EPI normalisation into MNI T1 space (via transformation parameters)

2. as in (1), but including a DARTEL-derived T1 template as intermediary

I ran into an issue with T1 segmentation (standard SPM settings) when trying to perform the first protocol.

Specifically, it appears that in certain cortical areas the segmentation assigns a very low probability to grey matter tissues.

To illustrate this, I used mricrogl to create 3D renderings of several single subject grey matter probability masks (obtained through SPM segmentation).

The first attached figure (fig.1) shows how this effect is consistent across subjects at different time points, and even when using a different T1 protocol on the same scanner.

Although not attached, I observed the same effect on T2 images.

Notably, in this study we don’t have field map information available (nor EPI distortion correction).

It can also be visually appreciated on T1 images that the affected cortical areas have low intensity contrast between grey and white matter

Further, I tried to optimise the segmentation by adjusting the different settings in the segmentation menu of SPM.

The best result I was able to achieve so far is shown in fig.2. 

While improved, it’s still noticeable that the areas are not optimally segmented.

This is further corroborated by binarising these images at a 25% intensity threshold, which still leaves some dark areas (i.e. holes or very thin areas).

On a last note, I have also tried segmentation in FSL, which was not improved after a brief exploration of the effect of different settings.

Freesurfer did however provide better segmentation, but I could only figure out how to get binarised images, not probability maps.

1) Has anyone ever come across similar issues with SPM segmentation and do they have advise on how to improve this?

2) Am I over-concerned and are the segmentations presented in fig. 2 of sufficient quality to proceed with the pre-processing? 

3) Are these images of sufficient quality to create a DARTEL template? (Particularly important as ideally I am trying to recreate a protocol that includes this)

4) Do I understand the combined SPM segmentation/normalisation procedure well in that a proper estimation of tissue probability maps is important in order to achieve good normalisation?

In essence, are my worries grounded or is the effect on normalisation not that strong?

5) If the current segmentations are not of sufficient quality, does anyone have experience with augmentation other software (eg freesurfer) in order to obtain tissue probability maps and transformations parameters which can be adapted in SPM?

Thank you very much for you time.

Sincerely,

Michaël

[log in to unmask]>" alt="Screen Shot 2018-12-06 at 16.50.32.png" width="1069" apple-inline="yes" class="Apple-web-attachment Apple-edge-to-edge-visual-media Singleton" src="blob:https://mail.uantwerpen.be/e78c821a-0e32-4de3-abf5-0ac7d179ef41">

[log in to unmask]>" alt="Screen Shot 2018-12-06 at 16.50.44.png" width="1069" apple-inline="yes" class="Apple-web-attachment Apple-edge-to-edge-visual-media Singleton" src="blob:https://mail.uantwerpen.be/a214e178-faf3-4042-a9a7-01b4de7ddef1">

[log in to unmask]>" alt="Screen Shot 2018-12-06 at 16.51.04.png" width="1070" apple-inline="yes" class="Apple-web-attachment Apple-edge-to-edge-visual-media Singleton" src="blob:https://mail.uantwerpen.be/bfa39320-f590-46bd-a974-286e73a6dcb3">