Oral Abstract 1: Inline Automated Post-processing and On-scanner Diffusion Tensor Maps Visualization for Cardiac Diffusion Tensor Imaging Using FIRE

Spin-echo Cardiac diffusion tensor imaging (cDTI)¹ is signal-to-noise ratio (SNR) limited, in part due to the long motion-compensated (velocity and acceleration, M₁+M₂) diffusion encoding gradient waveforms that extend echo times (TEs). Ultra-high-performance MRI systems (e.g. Gmax=200mT/m, Smax=200T/m/s) are a great opportunity for cDTI to minimize TEs and increase SNR; however, peripheral nerve stimulation (PNS) from slewing to higher gradient thresholds needs to be managed. Gradient waveform optimization, such as the open-source Gradient Optimization toolbox (GrOpt² ), provides more time-efficient gradient waveforms while managing constraints.  

The wider adoption of cDTI and SNR-efficient optimized gradient waveform approaches are limited by specialized sequence development in proprietary vendor software. Pulseq³ is another open-source toolbox that enables vendor-neutral pulse sequence programming in Matlab or Python,⁴ lowering the barrier for design.

The purpose of this work is to demonstrate an open-source approach to spin-echo cDTI data collection using a combination of GrOpt and Pulseq.

Methods:

In simulation, the minimum TE achievable for a spin-echo cDTI sequence was simulated for an ultra-high-performance system (G_max=200mT/m) and commodity system(G_max=45mT/m) with conventional trapezoidal waveforms (TRAP) and GrOpt waveforms. The following parameters were used:  b-value=[250, 500, 750, 1000] s/mm², RF pulse durations of T₉₀ = 3ms, T₁₈₀ = 5ms, gradient moment nulling = [M₁+M₂], time to TE=[10, 20, 30, 40, 50] ms, PNS threshold⁵ (PNS_thresh)=0.8. TE differences (∆TE_min) and percent signal gain of myocardium (T₂=46ms) were reported.

A proof-of-concept, ECG-triggered, M₁+M₂ compensated, single-shot spin-echo diffusion echo-planar imaging (EPI) sequence (ramp-sampled⁶) was implemented in Pulseq with GrOpt waveforms (Table-1 and Fig-2). The phase-encoding (PE) blip gradient was turned off for the first excitation to acquire a reference to correct for odd and even PE line shifts. Two non-diffusion-weighted images were then acquired with dual polarities for linear phase correction⁶ followed by diffusion-weighted images. A separate two-shot EPI calibration scan was acquired right before the imaging scans. Image reconstruction included ramp sampling interpolation, ghost correction, GRAPPA reconstruction, and homodyne Partial Fourier reconstruction. Mid-systolic imaging was targeted based on the ECG signal and localizer images. Two imaging protocols were tested (Table-1).

Results: In simulation, GrOpt consistently maximized hardware usage across all b-values and readouts, outperforming conventional trapezoidal (TRAP) waveforms. TE reductions over 50 ms led to estimated signal gains up to 200% (Fig-1). We integrated GrOpt into a single-shot EPI Pulseq sequence (Fig-2). Preliminary images show successful diffusion-weighting, with potential improvements in fat saturation and trigger delay selection.

Conclusion: GrOpt utilizes dynamic slewing to limit PNS while reducing TE durations. Our proof-of-concept sequence development and data acquisition shows promise of optimized open-source cDTI. Future work will include sequence design improvements (integrating reference and calibration scans, improved fat saturation) and exploring simultaneous ECG-triggering and respiratory navigating to allow for free-breathing acquisitions.
{

Funding: AHA.23PRE1018442, R01 HL152256, NSF 2205103, and ISMRM Research Exchange}