1357591 - Ferumoxytol-Enhanced 5D ROCK-MUSIC with Low-rank Tensor Reconstruction in Pediatric Congenital Heart Disease

ROCK-MUSIC is a 4D imaging pulse sequence that when used with ferumoxytol enhancement, shows great promise in pediatric congenital heart disease (CHD)¹. ROCK-MUSIC employs incoherent, rotating Cartesian k-space sampling and combined compressed sensing and parallel imaging reconstruction (ESPIRiT) for multiphase, whole-heart 3D imaging.^{1 3} As originally implemented, ROCK-MUSIC is prone to unpredictable aliasing that may manifest as flickering in some reconstructed frames. Recent developments in low-rank tensor decomposition for multidimensional image reconstruction can resolve cardiorespiratory motion by allowing for the separation of cardiac and respiratory phases.² We aim to evaluate the feasibility of using model-based low-rank tensor image reconstruction with ROCK-MUSIC to enable cardiac and respiratory motion resolved 5D-CMR with high temporal resolution and less aliasing artifacts.

Methods: We obtained retrospective data from six pediatric patients (age range, 2 days to 4 years) with known CHD. The patients were scanned on a clinical 3.0T magnet (Magnetom TIM Trio®, Siemens Medical) using the ROCK-MUSIC pulse sequence ([TE/TR] = 1.2ms/2.9ms, 0.8-1.1mm³ isotropic resolution, TA = 4.35-6.12min, flip angle=19-25°) during the steady-state intravascular distribution of ferumoxytol.¹ We applied the low-rank tensor method on all six and further performed ESPIRiT on a subset of three patients and qualitatively evaluated the images. We quantified the left ventricular end-diastolic volume (EDV), end-systolic volume (ESV), and left ventricular ejection fraction (LVEF) derived from each image reconstruction method.

Results: For each dataset, low-rank tensor decomposition generated a 5D ROCK-MUSIC  reconstruction consisting of 6 respiratory X 20 cardiac = 120 cardiorespiratory phases. Nine cardiac phases were generated from ESPIRiT. Fig. 1 provides an illustrative comparison of the image quality in one respiratory phase whereas Fig. 2 shows a comparison of image quality across several phases for both reconstruction methods. In this limited dataset, the overall image quality ranged between 3-4, with 4 having sharp diaphragmatic and cardiac border definition; the low-rank tensor reconstruction was effective at removing aliasing and respiratory motion artifacts while offering higher temporal resolution. Fig. 3 shows the cine images from multiple views for low-rank reconstruction. The mean bias for LVEF, EDV, ESV between the two reconstruction methods for this pilot dataset were 0.7%, 0.44ml and 0.38ml respectively.

Conclusion:

In this preliminary work, low-rank tensor reconstruction is feasible for extending ROCK-MUSIC to 5D-CMR by enabling cardiac and respiratory phase-resolved image reconstruction. Additional comparisons between ESPIRiT and low-rank tensor reconstruction in a larger dataset are needed to fully appreciate the performance of both approaches.