Hello all,
First of all, I’m new to Neurostars so apologies if this is not formatted quite as desired – happy to follow up with additional details.
For my research project we are processing three pre-existing structural MRI datasets, all of which have T1 and FLAIR images. The goal is creating maps of white matter hyperintensities in MNI standard space for each participant to use for later steps in the analysis. However, while our processing pipeline works well for two of the datasets, for the third dataset we are getting an artifact for many participants in which their lesion map implausibly shows white matter hyperintensities at the most posterior part of the brain, strong enough that they are outweighing all their other white matter hyperintensities. After attempting a few possible solutions, we’ve been unable to fully resolve this so I wanted to check if others have any insight into this.
I will briefly outline our processing steps here and can give more detail if needed. To preprocess the raw T1 and FLAIR images (.nii format), we rotated and cropped them in FSL, and then performed bias-field correction on the images using FAST in FSL. We then apply the Lesion Prediction Algorithm in the Lesion Segmentation Tool for SPM in MATLAB to the FLAIR images to create a map of white matter hyperintensities for each participant. Next, to register the lesion maps to MNI standard space, we take the T1 and FLAIR images (that went through the preprocessing above) and skull-strip them using HD-BET (whereas the LPA takes non-skull stripped images). We then use ANTS to register the skull-stripped T1 and FLAIR images, with the following three steps: 1) register the participant’s T1 to an MNI template image, 2) register the FLAIR to the T1, and 3) register the lesion map to the FLAIR.
As I mentioned, this all works well for two of the datasets, but the artifact of implausible lesions at the very back of the brain is appearing for the third dataset. Before we added bias-field correction to our pipeline, the artifact was appearing somewhat at the LPA stage (creating the lesion maps) and a lot after registering the lesion map to MNI space. (The images for that dataset were brighter in more posterior regions, probably due to something about their scanner.) Now that we have bias-field correction, the artifact is not appearing when we run the LPA, but it is still appearing for many participants after registration to the MNI template. It is significantly less strong than when bias-field correction hadn’t been added, but still strong enough to greatly throw off later analyses for that dataset, to the point that those participants cannot be used.
We have been stuck trying to resolve this for a while now and it would be great if anyone has suggestions as to what may be causing this and how we could modify our processing pipeline to prevent it occurring.
Thanks very much!
