Background

Earlier this year we learned that some anatomical T1 data were collected with GE’s GradWarp processing disabled.  GradWarp is GE’s method for correcting spatial image distortions that are introduced by physical gradient fields not being perfectly linear.  In practice, these gradient fields are nearly linear over a normal head volume but deviate at the edge of the field of view (FOV).

GradWarp was unintentionally disabled in the cni_efgre3d sequence during a software update on 10/21/2021.  The method was re-enabled during another update on 2/14/2022. Disabling GradWarp had a significant effect only on scans that were acquired with a rectangular (not square) FOV.  The reason is that GE combines the image resampling that corrects for the rectangular FOV together with the GradWarp processing.  Thus,  rectangular FOV scans were reconstructed incorrectly assuming equal FOVs in both X & Y.  For example,  in a case where the yFOV is only 80% of the xFOV, the final reconstruction will be stretched by 25% in the Y direction.  This is easily detected.

Processing

We have been in contact with all groups that acquired data with rectangular FOVs, and we worked with them to identify and fix the problem. We just completed rolling out a correction that implements the GradWarp processing retrospectively to all the relevant data on Flywheel.

We also applied the GradWarp correction to data acquired with a square FOV (most anatomical data is acquired this way). In this case the correction is quite small (the gradients are close to linear over the head).  The typical displacement for a pixel at the edge of the brain will be 1 mm or less  This is shown in the images below where white gridlines are pictured in gradient space (which appear as a perfect grid spaced every 10 pixels) while the pink gridlines are as they appear after GradWarp is applied.  In this case the underlying brain image was acquired at a resolution 0f 0.9 x 0.9 mm. The GradWarp spatial correction depends on the position — positions further from the gradient isocenter have larger deviations from linear and the GradWarp corrections are larger.

Identifying Updated Data

In a recent update, Flywheel added version control to user data.  This is described by Flywheel in their online document: File Versioning. We used this feature when correcting the DICOM and NIfTI for anatomical T1 data.  Specifically, you will see a number “2″ next to the filename (see below).  When you view, download or operate on the Flywheel data, by default you will be provided the GradWarp corrected data (version 2).  You can restore the original data at any time by clicking on the version control number “2″.  A popup window will appear and you can restore version “1″ as the default (note that this will create a new version “3″ of the file).

In addition, Projects with acquisitions containing uncorrected anatomical T1 data (not GradWarp corrected) have a CSV file attached to the project level. The files are named using the following convention: 

    cni-gradwarp_<group_name>_<project_name>.csv

This file contains information describing the acquisitions with uncorrected and subsequently fixed T1 data. The csv file contains the following information for each corrected scan: Group|Project|Subject|Session|Acquisition|AcquisitionId. This information can be used to help you find data that was corrected.