Undersampled 3D T1 MPRAGE: Diagnosing a Parallel-Imaging Artifact
A research protocol uses a 32-channel head coil at 3 T with a 3D MPRAGE accelerated 4-fold (2x2) in the phase and partition directions. Reconstructions show periodic banding and 'ghosting' replicas in the partition direction, most prominent at the skull base and posterior fossa. The technologist suspects the coil, but swapping coils does not change the pattern. The acceleration factor was recently increased from 3 to 4 without changing the calibration scan.
1. What is the most likely cause of the banding, and how would you confirm it?
Read the worked answer
The pattern - periodic replicas in the partition (slice-encoding) direction, worst where B0 inhomogeneity is greatest (skull base, posterior fossa) - is classic for insufficient or mismatched coil sensitivity calibration in a parallel-imaging reconstruction (GRAPPA/ARC or SENSE). At 4-fold acceleration the reconstruction relies more heavily on accurate sensitivity maps; if the calibration scan was acquired at a different contrast, resolution, or time than the accelerated data, the maps are wrong and residual aliasing appears.
Confirmation steps:
- Reconstruct with no acceleration (or 2-fold) using the same raw data; if the banding disappears, the acceleration/calibration is implicated.
- Check the calibration scan: was it acquired with the same coil, same FOV, same orientation, and a comparable contrast? For MPRAGE, an in-line calibration using the same sequence is preferred over a separate low-resolution reference.
- Inspect coil sensitivity maps for noise or dropout, especially near the skull base where B1 is inhomogeneous.
- Verify the g-factor map if available; high g-factor regions indicate poor coil geometry for that acceleration.
Swapping coils not changing the pattern argues against a hardware fault and toward a calibration or reconstruction issue.
2. How would you redesign the acceleration to preserve scan time while reducing the artifact?
Read the worked answer
Options, in order of practicality:
- Re-acquire the calibration scan with the same sequence and contrast as the accelerated data (in-line calibration, matching FOV and orientation). This is the most common fix.
- Redistribute acceleration: instead of 2x2, use 3-fold in the phase direction and 1x in the partition direction, or use a CAIPIRINHA-style pattern with phase shifts between partitions to spread aliasing and improve g-factor. Partition-direction acceleration is more prone to aliasing near the skull base because of B0 inhomogeneity.
- Use a larger calibration kernel or a split calibration with more reference lines.
- If time allows, drop to 3-fold total acceleration (which previously worked) and compensate with a slightly longer TR or fewer averages.
- Consider compressed sensing or wave-CAIPI if the vendor supports it, which tolerates higher acceleration with better artifact behavior.
After redesign, inspect the g-factor map and compare SNR and artifact level at the skull base before accepting the protocol.
3. The banding persists only in the posterior fossa even after recalibration. What non-parallel-imaging mechanisms could contribute, and how would you test them?
Read the worked answer
Residual posterior-fossa artifact can come from:
- B0 inhomogeneity and susceptibility near the petrous bone and air-tissue interfaces, causing local off-resonance that degrades sensitivity estimation and creates pile-up. Test with a B0 field map and compare artifact location to the field map; try higher-order shimming or a localized shim volume over the posterior fossa.
- Motion between calibration and accelerated data (swallowing, CSF pulsation) causing inconsistent sensitivity maps. Test with a repeat acquisition and motion monitoring; use in-line calibration to minimize the time gap.
- Gradient nonlinearity and eddy currents interacting with the partition encoding. Test by comparing a non-accelerated reference and checking whether artifact moves with gradient settings.
- RF transmit field (B1+) inhomogeneity at 3 T causing contrast variation that the reconstruction misinterprets. Test with a B1 map and consider dielectric pads or parallel transmit.
Systematically eliminate each by acquiring a non-accelerated reference, a field map, and a B1 map, and by comparing artifact location across them.