Referenceless Ghost Correction for Simultaneous Multi-Slice Cardiac Diffusion Tensor Imaging (SMS cDTI)

Background: Cardiac diffusion tensor imaging(cDTI) with single-shot EPI readout enables noninvasive myocardial microstructureassessment, but long acquisitions and breath holds limit its clinical translation¹. To improve efficiency, simultaneous multi-slice(SMS) imaging with parallel imaging has been introduced². Although SMS increases efficiency and reducing the overall number of breathholds required, it may extend the duration of every breathhold because two-step SMS requires separate SMS and PI calibration scans, in addition to the phase correction navigator scans EPI readouts rely on to address Nyquist ghost artefacts from odd–even k-space misalignments. These extra scans not only increase the number of required heartbeats and prolong breath holds, but can also be misaligned with the imaging data because of cardiac and respiratory motion. Building on our previous work on referenceless EPI ghost correction for single-band(SB) cDTI³ and one-step SMS reconstruction^4,5, we aim to minimize additional calibration and correction scans to deliver SMS with the same breathhold duration as SB scans, in the heartbeats of an SB scan with reduced overall number of breathholds and preserved accuracy.

Methods: Twenty healthy volunteers were scanned on a 3T(Vida, Siemens) using SMS-STEAM-EPI with respiratory navigator guidance⁶. Protocol:b=30,150,600s/mm²;slice thickness=8mm;in-plane2.8x2.8mm²;SMS=2;PI = 2. SMS and PI calibration data were obtained. SB data were acquired with the same protocol.

All data and calibrations underwent offline referenceless first-order ghost correction via a search over linear phase slope and offset. The ghost/objective minimization⁷ was embedded within the reconstruction, evaluating the cost on the unaliased image at each iteration.

SMS data were reconstructed offline using our in-house Matlab implementation of 2-step and 1-step reconstruction. SB data were reconstructed using GRAPPA. cDTI data were analysed using our in-house post-processing tool⁸.

Results: Figures 1-2 show apical and basal slices at three b-values and diffusion maps for SB, 2-step and 1-step SMS with referenceless correction. Ghost and leakage artefacts were reduced with 2-step and 1-step SMS, but the 2-step SMS shows more residual artefacts than the 1-step SMS. The 1-step SMS maps were closest to the SB maps.
MD, FA, and HA showed no significant method difference. Only |E2A| showed a slice effect in SB (basal >apical, p=0.0007), consistent with heart rate variation between separate acquisitions. In contrast, SMS acquires both slices simultaneously and showed no apical versus basal difference (p_2step=0.27,p_1step=0.17)(Fig. 3).

Conclusion: We successfully showed the effectiveness of referenceless ghost correction for SMS cDTI, as well as the improved image quality and scanning efficiency of 1-step SMS reconstruction compared to 2-step. These support a practical protocol with fewer scans and robust myocardial microstructure quantification, paving the way for the translation into the clinic.

Figure 1. Apical and basal diffusion-weighted images (b₀ = 30, b = 150, b = 600 s·mm²) generated from SB, two-step SMS, and one-step SMS, all of them with referenceless EPI ghost correction. Two-step SMS exhibits residual Nyquist ghosts and inter-slice leakage (arrows), whereas one-step SMS suppresses both artefacts.
Figure 2. Example diffusion maps of mean diffusivity (MD), fractional anisotropy (FA), helix angle (HA), and absolute secondary vector angulation (|E2A|) for apical and basal slices from SB, two-step SMS, and one step SMS processed with referenceless EPI ghost correction. Numeric values indicate slice-wise mean ± SD within the myocardium. One-step SMS is closest to SB across slices.
Figure 3. Statistical comparison across 20 volunteers. Violin plots show distributions of MD, FA, HA, and |E2A| for SB, two-step SMS, and one-step SMS with referenceless ghost correction. Significant differences (p < 0.05) are annotated. Metrics were broadly similar across methods; no significant differences were observed except for basal |E2A|, which showed a method effect (pairwise p = 0.017 and p = 0.008). This however is likely a limitation of SB, as apical and basal slices needed to be acquired separately, which due to variations in heart rate likely led to different timepoints in the cardiac cycle and therefore significantly different E2A values (p0.0007). The SMS datasets appears more consistent, and no significant differences are observed between 2-step and 1-step apical and basal slices (p2step=0.27, p1step=0.17).