Print

Print
I am new to fMRI analysis and I am trying to preprocess resting state EPI images from a GE Discovery 3T scanner with distortions in the orbitofrontal cortex, leading to a failure of normalization using T1 unified segmentation. My guess is because of bad coregistration due to the shape miss-match between the images.

The suggestion I received from the SPM mailing list is to segment the mean EPI and use the generated segmentation parameters to normalize my EPI images, which produces images in the shape of the MNI templates, but my question is this; is this method producing images that are correctly morphed so that functional connectivity can be analyzed or are the images simply stretched to fit the shape of the MNI template?

It is my understanding that this method of directly segmenting the EPI mean requires good contrast in the EPI images. My image resolution is 2.25x2.25x3mm (288mmx288mm FOV). The estimates I'v read state that resolutions of 3mm or over are too bad for this approach. (I'v attatched a screenshot of my EPI data for anyone wanting to judge the contrast)

My current preprocessing strategy would thus be the following:

1. temporally&spatially align EPI iamges
(2.) Coregister T1 to EPI.mean
(3.) manually reorient T1, EPIs and EPI.mean to better match template
4. segment EPI.mean
5. normalize EPIs using EPI.mean.seg.mat
6. Segment T1
7. normalize T1 using T1.seg.mat
8. smoothe T1 and EPI images

(The end goal of all this is to use the T1 to define WM and CSF ROIs to use as regressors when modeling the activation)

My final question is wether steps 2 and 3 are necessary at all. My reasoning is that by coregistering the EPIs and the T1, the T1 can be used to reorient the EPIs, wherein it is hard to find the AC. How much does this improve the accuracy of the normalization step?

-------------------------------
Andreas Lidström, Research Assistant
Dept of Women’s and Children’s Health
and Neurology Clinic,Karolinska Hospital,Sweden