1351465 - Motion-Resolved, Non-ECG gated, Free-Breathing 3D Low-Rank Tensor Based Reconstruction for Hemorrhagic Myocardial Infarction Characterization

Hemorrhagic myocardial infarction (hMI) has emerged as the most serious form of tissue injury heart attack patients [1, 2]. CMR is the gold standard for characterizing hMI but it is time consuming, affected by motions, and creates discomfort to the patients as the exams are run with multiple breathholds and ECG-gating. We studied whether it is possible to integrate multiple measurements into a single 15-min free-breathing, non-ECG, motion-resolved whole LV 3D acquisition using low-rank tensor-based reconstruction, with the goal of determining LVEF, microvascular injury (microvascular obstruction (MVO) and hemorrhage) and myocardial infarction size based on LGE.

Methods:

Sequence

Design: A free-running 3D Cartesian sequence with IR preparation and multi-echo GRE readout was used to collect interleaved navigator and imaging training data (Fig. 1). IR-preparation is used to maximize contrast between MI and remote myocardium.
Image Model: The images were represented as a 6-way tensor χ with a combined spatial dimension and 5 time dimensions (T1 recovery, T2* weighting, cardiac motion, respiratory motion and contrast dynamics). To utilize the anatomical similarity during contrast progression, we jointly reconstructed all images using the multitasking framework [3-5]. The Post-contrast acquisition was reconstructed in a time-resolved fashion to identify late gadolinium enhancement images during a steady-state. An ICA based algorithm was used to identify 4 respiratory and 24 cardiac phases. Basis functions for each state were estimated from the subspace training data using Bloch-constrained low-rank tensor completion.

Data acquisition: MIs were created in dogs (n=3) and imaged at weeks 1 and 8 post MI. Scan parameters were: TR=13.20 ms, 6 TEs/DTE=1.42/2.01 ms, recovery period=2520ms (scan time: < 15 mins for all subjects). Reference 2D cine, T2* and 2D LGE images were acquired at diastole during end-expiration breath-holds. All images were acquired in 1.4x1.4x6 mm³.
Data Analysis: LVEF was computed using standard and 3D LRT cine images. MI size and transmurality and microvascular obstruction (MVO) were determined from LGE images. Size of hemorrhage was determined from standard and LRT-based T2* weighted images (TE=11.42 ms). Paired t-tests and linear regressions were used to determine differences between conventional and LRT approaches.

Results:

Fig. 2 shows typical T2*-w and LGE images obtained using conventional (ECG-gated, breath held) and LRT based images. There were no statistical differences between LVEF, size of hemorrhage, extent of MVO and MI size between the approaches (p >0.2). There was strong correlations between MI transmurality and size determined using the same approaches.

Conclusion: The proposed LRT approach shows early promise for accurate characterization of hMI without respiratory and cardiac gating within 15 minutes. Additional studies are needed to assess the robustness of the approach and to assess clinical feasibility.