Question
Question Posted 03/27/26:
Hello,
In the documentation [https://adni.loni.usc.edu/data-samples/adni-data/neuroimaging/mri/mri-pre-processing/] it is mentionned that N4 bias field correction may be applied on T1 images from ADNI3 and ADNI4 (with scanner's internal preprocessing). We wish to perform an analysis on images across the different ADNI datasets. However for previous releases of ADNI, N3(m) bias field correction has been applied during the pre-processing of images. Should N4 bias field correction be applied on the fully pre-processed images? Or for these images should the version before N3(m) correction (with gradwrap and B1 correction only) be preferred? Thanks for your help
Hello,
In the documentation [https://adni.loni.usc.edu/data-samples/adni-data/neuroimaging/mri/mri-pre-processing/] it is mentionned that N4 bias field correction may be applied on T1 images from ADNI3 and ADNI4 (with scanner's internal preprocessing). We wish to perform an analysis on images across the different ADNI datasets. However for previous releases of ADNI, N3(m) bias field correction has been applied during the pre-processing of images. Should N4 bias field correction be applied on the fully pre-processed images? Or for these images should the version before N3(m) correction (with gradwrap and B1 correction only) be preferred? Thanks for your help
Response posted 03/30/26 by ADNI MRI Core:
Thank you for your question to the ADNI MRI Core.
After checking with our experts they confirmed that ADNI has never used N4 in any preprocessing of data.
Separating the issues:
Gradient distortion correction (grad warp): Typically, MR images are reconstructed from k-space assuming that the imaging gradients are linear across the field of view. In practice they are not. This introduces a warping in the reconstructed image. Current scanners can correct for this using on-board corrections. In ADNI-1 Philips performed this correction for all three imaging gradients; GE did it for two ; and Siemens could do none or two directions. We unwrapped the images as needed. In ADNI-2 the scanner fleet shifted and fewer images needed to be unwrapped. In ADNI-3 (nearly?) all scanners could perform 3D correction. In ADNI-4 all scanners have 3D correction. Short answer : If there is a GW version of an image it was created in post-processing, if not then 3D distortion correction should have been done on the scanner.
B1-inhomogeneity correction: In ADNI-1 Philips had CLEAR; some Siemens scanners may have had “pre-scan normalization” but many did not; and some GE scanners had PURE. They are all basically the same idea: acquire images with the head coil that has a shading artifact and acquire an image with the body coil (that is assumed to have a homogenous response across the field of view) and use that to correct all the other head coil images. In ADNI-1 this step was done in post-processing as necessary. On top of that correction for ADNI-1 the N3 bias correction was applied followed by the bias correction from spm5 segmentation. In ADNI-2 and beyond this was dropped and users are expected to deal with bias (shading artifact) correction as they see fit.
Unsure of what particular data sets the user is attempting to analyze. I would always use the GW version if available (if it exists it was needed).
Whether to use additional corrections like B1-ratio and/or N3m probably depends on the analysis being performed and if it handles shading artifacts.
ADNI MRI Core
After checking with our experts they confirmed that ADNI has never used N4 in any preprocessing of data.
Separating the issues:
Gradient distortion correction (grad warp): Typically, MR images are reconstructed from k-space assuming that the imaging gradients are linear across the field of view. In practice they are not. This introduces a warping in the reconstructed image. Current scanners can correct for this using on-board corrections. In ADNI-1 Philips performed this correction for all three imaging gradients; GE did it for two ; and Siemens could do none or two directions. We unwrapped the images as needed. In ADNI-2 the scanner fleet shifted and fewer images needed to be unwrapped. In ADNI-3 (nearly?) all scanners could perform 3D correction. In ADNI-4 all scanners have 3D correction. Short answer : If there is a GW version of an image it was created in post-processing, if not then 3D distortion correction should have been done on the scanner.
B1-inhomogeneity correction: In ADNI-1 Philips had CLEAR; some Siemens scanners may have had “pre-scan normalization” but many did not; and some GE scanners had PURE. They are all basically the same idea: acquire images with the head coil that has a shading artifact and acquire an image with the body coil (that is assumed to have a homogenous response across the field of view) and use that to correct all the other head coil images. In ADNI-1 this step was done in post-processing as necessary. On top of that correction for ADNI-1 the N3 bias correction was applied followed by the bias correction from spm5 segmentation. In ADNI-2 and beyond this was dropped and users are expected to deal with bias (shading artifact) correction as they see fit.
Unsure of what particular data sets the user is attempting to analyze. I would always use the GW version if available (if it exists it was needed).
Whether to use additional corrections like B1-ratio and/or N3m probably depends on the analysis being performed and if it handles shading artifacts.
ADNI MRI Core
