To quantitatively assess and compare the effectiveness of three reverse-gradient method susceptibility artefact correction tools in the correction of spin-echo (SE) and gradient-echo (GE) EPI images of the brain acquired using a dual echo EPI sequence. A secondary objective was to measure the effect of pixel bandwidth, SENSE factor and slice thickness on artefact correction.

Five healthy volunteers were scanned using a 3T Philips Achieva MRI system with an 8-channel SENSE head coil. A T1-weighted sagittal whole brain scan was acquired, which was used as an anatomical reference for co-registration. Multiple GE-EPI and SE-EPI images were acquired for each patient using a single-shot dual-echo EPI sequence with the following parameters: TR = 1500ms, TE(GE) = 35ms, TE(SE)=120ms, FOV = 192mm x 192mm, Echo Train length (ETL) = 62.

The following parameters were varied:

Pixel bandwidth: 636Hz, 893Hz, 1861Hz

Slice thickness: 2mm, 4mm, 6mm

SENSE factor: 1, 2 3

For each set of parameters, the EPI sequence was run twice, with opposite polarity of the phase-encoding gradient. The acquired images were then corrected using three correction tools; EPIC¹, FSL TOPUP², and SPM HySCO³. These tools employ the reverse gradient method of susceptibility correction, whereby two images acquired with opposite polarity of the phase-encoding gradient (therefore having pixel displacements in opposite directions) are used to calculate a pixel displacement map (Figure 1). This map can then be used to correct image series acquired in the same way. The correction tools were used with their default settings, as these were considered representative of typical usage. The corrected images were then independently co-registered with the high resolution T1 weighted scan and the maximal value of normalized mutual information (NMI) was recorded⁴. NMI is a measure of geometric similarity between two images, and was therefore used to determine the effectiveness of the correction of the geometric distortions⁵. Mann-Whitney and Kruskal-Wallis non-parametric tests were used to determine significant differences in NMI with a confidence value of p<0.05.

The change in mutual information after correction was found to be significant only for the EPIC correction method, according to the Mann-Whitney u-test (Figure 2). We found that this was the case for both SE-EPI and GE-EPI images. The difference in NMI gain using the EPIC correction tool on SE-EPI images with varying bandwidth was found to be significant (Figure 3). It was generally observed that susceptibility artefact correction reduced the variation in NMI between images of different pixel bandwidth, slice thickness, and SENSE factor.The EPIC correction tool achieved the most significant gains in mutual information for both GE-EPI and SE-EPI images. Upon inspection, EPIC correction resulted in images with fewer artefacts and noticeable improvements in geometric distortions (Figure 4). The HySCO correction performed poorly in this study, and did not satisfactorily reverse the geometric distortions. TOPUP corrected the geometric distortions, but was found to introduce banding artefacts. It should be noted that parameter optimization might improve the performance of these tools; our experience with these settings suggests that this may be the case, however due to the large number of free parameters, an investigation of the effects of each of them was outwith the scope of this study. The effect of pixel bandwidth on the image correction using EPIC was found to be significant (Figure 5); however the effects of SENSE factor and slice thickness on the correction quality were less apparent. Generally, these tools were equally effective in the correction of geometric distortions in both GE-EPI and SE-EPI images.EPIC was found to be the most effective of the three correction tools in the present study, both in terms of the gain in NMI and with respect to the reduction of visible artefacts in the correction of both GE-EPI and SE-EPI images. Furthermore, more mutual information was recovered at lower pixel bandwidths, reducing the impact of acquisition bandwidth on the geometric accuracy of SE-EPI images post correction.