Enhanced Simultaneous Multi-Slice Diffusion Tensor Cardiovascular (SMS DT-CMR) using One-Step Extended SENSE with Interslice Leakage Constraints

Diffusion tensor cardiac magnetic resonance (DT-CMR) is a promising technique for non-invasive assessment of myocardial microstructure but is limited by long acquisition times and low signal-to-noise ratio (SNR) [1]. To accelerate these acquisitions, simultaneous multi-slice (SMS) imaging combined with in-plane parallel imaging (PI) is a promising method[2-3], but reconstruction methods are prone to inter-slice leakage artifacts, hindering myocardial microstructure quantification.  

We propose an extended one-step SENSE reconstruction framework for SMS DT-CMR that effectively addresses inter-slice leakage with k-space constraints, combining the advantages of both SENSE and GRAPPA approaches. By leveraging null space projection and Hankel-structured low-rank constraints in k-space, the technique suppresses inter-slice leakage while preserving signal integrity in the target slice. We also integrate image-domain regularization to further enhance the quality of SMS reconstruction. 



Methods:

We implemented an extended SENSE framework utilizing an extended controlled aliasing calibration for one-step reconstruction [4]. Sensitivity maps and null space operator were estimated from a single-band reference dataset. To control interslice leakage, we integrated null space consistency and a Hankel-structured low-rank prior into the SENSE framework. Additionally, we incorporated locally low-rank (LLR) regularization in the image domain to reduce noise in the diffusion images. 

SMS and single-band (SB) DT-CMR data were acquired at 3T (MAGNETOM Vida, Siemens) using single-shot EPI. SB porcine ex-vivo data were analyzed retrospectively, and in-vivo SMS data acquired with a multi-band STEAM sequence [5] were used for feasibility tests. Data underwent phase correction, SMS-PI reconstruction, and DT-CMR post-processing in MATLAB.  



Results:

Figure 1 compares the proposed one-step extended SENSE method with the conventional two-step GRAPPA [6] using simulated SMS data (calculated from SB data) at SMS=2, PI=2, displaying reconstructed images and their corresponding normalised root mean square error (nRMSE) maps and structural similarity index (SSIM) maps. The proposed method shows lower mean nRMSE, and higher mean SSIM at all b-values tested (b=0, 500 and 1000smm-2