What is parallel imaging? Describe in brief its technique.

Parallel imaging is an MRI acceleration method that uses the spatial sensitivity of multi-channel receiver coils to undersample k-space and reconstruct full–field-of-view images, thereby shortening acquisition time without changing contrast weighting. It achieves this by reducing phase-encoding steps by an acceleration factor RRR and resolving the resultant aliasing using coil sensitivity information in the image domain (SENSE-type) or k-space interpolation (GRAPPA-type) during reconstruction.

Definition

Parallel imaging leverages independently digitized signals from multiple receiver coil elements with distinct spatial sensitivity profiles to provide additional localization information beyond conventional phase encoding. By exploiting these sensitivities, fewer phase-encoding lines are acquired, decreasing scan time proportionally to the acceleration factor RRR within SNR and geometry constraints.

Core technique

Acquire undersampled multi-channel k-space with reduced phase-encoding lines by factor RRR (1D or 2D undersampling), accepting controlled aliasing in the reduced-FOV images.
Collect calibration data: either a separate low-resolution sensitivity map for image-domain methods or auto-calibration signal (ACS) lines embedded in k-space for k-space methods.
Reconstruct full-FOV images by “unfolding” aliasing using coil sensitivity maps (SENSE family) or by synthesizing missing k-space data using learned interpolation kernels (GRAPPA/SMASH/SPIRiT family).

Reconstruction methods

Aspect	SENSE (image-domain)	GRAPPA/SMASH/SPIRiT (k-space)
Core idea	Unfold aliased images using explicit coil sensitivity maps	Synthesize missing k-space lines using calibrated convolutional kernels across coils
Calibration	Requires accurate coil sensitivity estimation; separate reference or self-calibrated variants exist	Uses ACS lines; does not require explicit sensitivity maps
Pros	Direct control via regularization; efficient when maps are accurate	Robust to map errors; widely used in EPI and routine sequences
Cons	Sensitive to map errors and motion; FOV must match encoding	Kernel miscalibration can cause residual artifacts; ACS costs time

SNR and g-factor

Parallel imaging incurs an SNR penalty approximated by

$$ SNRR≈SNR0/(R g)\mathrm{SNR}_R \approx \mathrm{SNR}_0 / \big(\sqrt{R}\, g\big)SNRR≈SNR0/(Rg), $$

where ggg is the geometry factor describing noise amplification from coil geometry and conditioning of the reconstruction. Coil configuration, acceleration direction(s), and ACS quality strongly affect ggg and thus practical acceleration limits, with typical routine RRR values of 2–3 in 1D and modest 2D accelerations when coil coverage is favorable

Benefits and pitfalls

Benefits: shorter acquisitions, shorter breath-holds, fewer motion artifacts, and shorter echo trains that can mitigate distortions or allow higher resolution/coverage within time limits
Pitfalls: SNR loss and noise amplification (high ggg), residual aliasing from poor calibration or motion, and dependence on coil number/placement and acceleration direction

Brief “how-to” in practice

Choose acceleration factor and direction(s) aligned with coil sensitivity variation; set ACS lines if using k-space methods.sciencedirect+1
Acquire undersampled data with multi-channel coils and the selected calibration strategy (reference scan or ACS)
Reconstruct with SENSE-type (map-based unfolding) or GRAPPA-type (kernel-based interpolation), verifying artifact-free images and acceptable ggg-factor/SNR.