TH-CD-207-11: Gradient Nonlinearity Calibration and Correction for Head-Only Asymmetric Gradient System




Due to engineering limitations, the gradient fields in clinical MRI inevitably contain high-order, nonlinear components. The presence of gradient nonlinearity (GNL) causes image geometrical distortion. Standard correction methods are based on parameterization of the simulated gradient fields with spherical harmonic polynomials. Conventional whole-body gradient systems typically employ symmetric designs, and the GNL for such systems usually contains only odd-order terms (up to 5th-order). Recently, a high-performance, head-only gradient system was developed. Due to asymmetric design, this new system exhibits more complex GNL profiles. Here, we demonstrate measurement-based high-order(N>5) GNL-correction on this new system using a fiducial phantom and iterative model-fitting procedure.


The Alzheimer's Disease Neuroimaging Initiative (ADNI) phantom was scanned with a 3D IR-FSPGR sequence (sagittal acquisition plane, Nx=Ny=256, Nz=196, x= y=1.05mm, z=1.3mm, BW=125kHz) on the head-only gradient system operating at 80mT/m, 500T/m/s. This phantom contains 160 fiducial spheres (diameter=1.0 or 1.5cm) that are distributed within a diameter=20cm spherical shell. The spatial positions of the fiducials were then measured from the distorted images and fit using a spherical harmonic polynomial model and iterative fitting procedure. The model order was increased from 1 to 10 step-by-step to test the effects of including high-order terms (N>5). Then, the coefficients were used to correct distortion. The residual root-mean-square-error (RMSE) was calculated from fiducial positions tracked from GNL-corrected images.


The RMSE analysis shows that GNL-distortion is reduced from 3.93 to 0.39mm by using correction terms up to 7th-order including both even-and odd-order terms. The addition of higher orders (N>7) provides negligible benefit.


The GNL of a high-performance, head-only gradient system was successfully measured. The use of up to 7th-order correction terms was found to be sufficient for correcting GNL-distortion in a typical brain scan. These techniques could also improve geometric accuracy for whole-body MRI systems used for radiation therapy planning.