Quantifying changes in geometric distortion across a gradient coil replacement

Introduction: Gradient coils may be replaced on MRI systems as part of hardware upgrades or as a result of hardware failure. These events could introduce a step change in longitudinal quantitative neuro-imaging studies. Gradient winding errors cause even nominally identical gradient coils to differ in their field. Additionally the precise alignment of the gradient coil in the magnet and the active shielding coil can alter the gradient profile or eddy current effects, and hence geometric distortion. These considerations are increasingly important for short-bore and head-only gradient coils where the linearity is reduced. It is currently possible to detect pathological volumetric changes as small as 0.25% [1] with MRI and for such applications, maintaining consistent volumetric measurement across a gradient coil replacement is of crucial importance. To quantify the effect of a gradient coil replacement, geometric distortion was measured using a detailed 3D phantom before and after the event. The resulting distortion maps were compared, identifying regions where the change is negligible, and the potential impact on volumetric measures. Method: The gradient coils were replaced on a GE Signa 1.5T system after spike artifacts appeared in DTI sequences. The original 10-year old coils were replaced by a new set manufactured to the same design. A 3D phantom with high spatial definition was imaged 4 times on separate occasions before and after the replacement. The MR images (FSPGR; FOV=24×24×25.2cm; Matrix=256×256×252; bandwidth=968Hz/pixel; no distortion correction) were non-linearly registered to a CT reference image (assumed artifact free). A set of points around the magnetic field isocenter where the geometric distortion is low were used to estimate the global rigid-body transformation mapping the MR image to the reference image. The rigid-body component was removed from the non-linear transformation leaving a vector map of geometric distortion [2]. The distortion maps from the prereplacement time points were averaged (Vpre,mean) and subtracted from the post-replacement average (Vpost,mean), giving a vector field representing change in geometric distortion (Vdiff). An earlier reproducibility study estimated the precision of the method to be 0.3mm over 95% of the phantom volume. Based on this, a lower threshold of 0.3mm was applied to the magnitude of the vector field (|Vdiff|) to highlight areas of change. The standard error of the difference (SE(Vdiff)) was calculated using the expression n p p p SE post pre diff } ) ( ) ( { ) ( 2 2 σ σ + = where n is the group size (4), and σpre(p)