· researchers seeking a deeper understanding of EPI, associated limitations, typical artefacts and correction strategies

· recognise appearance and underlying physical mechanisms behind typical image artefacts associated with echo-planar imaging

· optimise protocol parameters to reduce the incidence of artefacts at source

· select appropriate remedial actions to correct for the residual imperfections in post-processing

It is instructive to first address the artefacts and limitations associated with traditional single-shot EPI without the use of parallel imaging, as the basic principles remain valid for other EPI variants. Through segmenting the trajectory or undersamping it is possible to scale down effects of the basic deteriorative phenomena, but they cannot be removed completely. Furthermore, both segmentation and parallel imaging introduce additional sources of errors, which may in their turn result in artefacts if not handled appropriately.

Single-shot EPI places stringent demands on the fidelity of the gradient system. However, even though it substantially stresses the gradient hardware the resulting k-space acquisitions are mostly too slow to ignore the signal evolution during the readout process. Both signal decay and undesired phase accumulation due to shimming imperfections contribute strongly to this evolution. Furthermore, strong oscillating gradients cause both complex eddy-currents in the conductive structures of the cryostat and non-linear phase accumulation due to the concomitant fields (so-called Maxwell terms). Some of the abovementioned phenomena (e.g. readout delays, fast eddy currents, etc.) result in periodic effects in k-space and therefore contribute to well-known EPI ghost artefacts. Other, such as signal decay or phase accumulation due to the resonance offsets, are characterised by a smooth evolution in k-space and cause more localized errors in the resulting images, such as regional blurring and geometric distortions.

Gradient-echo EPI shares a family of artefacts with other T2*-weighted sequences, which are associated with intra-voxel dephasing. Well-known signal drop-out a.k.a. void artefacts are oft seen in neurologic echo-planar images, especially in the pre-frontal areas, around ear canals or in basal slices. Due to the relatively low spatial resolution and increased slice thickness typical for single-shot techniques, signal void artefacts appear much exaggerated in EPI as compared to anatomical T2*-weighted imaging. EPI also adds its own flavour to the phenomenon, showing spatial contrast variations in the areas affected by local B0 gradients.

In spin-echo EPI intra-voxel dephasing occurring due to the shimming imperfections is refocused, which recovers signals also from the areas, where gradient-echo EPI would be void of signal. Although signal recovery may appear beneficial, such areas often present a problem due to the extreme geometric distortions, which may cause the dislocated signals to overlap with neighbouring distortion-free regions. Spin-echo EPI may also be affected by drop-out artefacts if the contrast preparation module induces substantial phase gradients within the image, as may occasionally happen for diffusion-weighted imaging in presence of physiologic motion. The resulting phase gradients may shift the echoes outside of the restricted k-space sampling window of EPI, which is associated with its relatively low resolution and oft is further reduced due to the partial Fourier sampling.

Severity of many of the EPI artefacts scales with the duration of the readout train. For that reason EPI sequence optimisation normally starts with shortening the readout gradient. Typically very high readout bandwidth is used, limited by either the maximum gradient strength or the SNR requirements. To minimize dead times ramp sampling is employed. In order to minimize the echo spacing maximum available slew rates are used. Oftentimes however, this is not achievable in vivo due to the possible violation of the peripheral nerve stimulation (PNS) safety limits. In this case a compromise between reducing the readout bandwidth and the slew rate is sought resulting in the minimum possible echo spacing. It is important to note, that PNS limits are anisotropic: for the head-first/supine subject positioning they are most restrictive in the Y direction and most relaxed in the X direction, respectively. Therefore X axis is often used to play out the rapidly oscillating readout gradient. Generally, as both eddy-current and electrical properties of the gradients are also anisotropic, it is oft recommended to use the pure X axis for the readout. Mixing in substantial contributions of the other gradient components (e.g. through the slice tilt or in-plane rotation) oft results in more complex trajectory deviations which may challenge ghost suppression algorithms.

Shortening the readout train, while marinating the spatial resolution, may also be achieved by reducing the number of echoes with a corresponding increase of the k-space step in the phase-encoding direction. The gaps in k-space are filled either by interleaving subsequent shots or using parallel imaging. Segmented EPI is highly susceptible to ghosting if inconsistencies between the shots cannot be suppressed or calibrated for. Such inconsistencies may occur due to subject motion, transient signal evolution or scanner drifts. Nowadays receiver arrays with a high number of channels have become routinely available, therefore parallel imaging is often preferred to interleaving for its insensitivity to errors between shots and a shorter minimal imaging time. However, EPI presents a special challenge to parallel imaging as well. Indeed, as the echo-planar images always suffer from some degree of geometric distortions, coil calibration data should ideally match these as well. One may consider segmented EPI with an equivalent readout duration as a natural choice. However, if coil calibration data are affected by ghosting, the parallel imaging reconstruction “learns” to reproduce these artefacts throughout the entire time series.

Presented above is a brief overview of the EPI artefacts along with some background information on their origins. In the presentation the audience will be invited to develop an appreciation of the whole complex of effects associated with EPI artefacts as well as the understanding why and when no perfect correction is possible. Thereafter a review of the established remedial approaches will be provided along with the current research trends.