1349571 - Autonomous radial k-space sampling in 5D

Pre-defined k-space acquisition schemes, such as golden angle-based approaches, can be cardio-respiratory binned for multi-shot reconstruction.¹ However, this strategy suffers from suboptimal distribution of samples in k-space and is time-consuming. In 2D, it has been shown that an autonomous trajectory, adaptive radial k-space sampling approach (ARKS), can improve multi-shot sampling uniformity by constantly monitoring and optimizing the acquisition of k-space data in a closed-loop.² Here, we extended ARKS to 5D imaging by performing 3D spatial acquisitions with cardiac and respiratory binning using physiologic signals acquired in pediatric patients.

Methods:

Cardiac and respiratory signals from 15 pediatric patients (average age 12.1±5.8 years old) were extracted from the 4D-flow scans and binned into 10 and 3 phases, respectively. IRB-approved waiver of informed consent was granted to perform this study. Four patients were under anesthesia.

A 3D predetermined, GA-based spiral phyllotaxis bSSFP sequence was simulated with the following parameters: TR 3.1msec, 10 ramp-up lines, 22 spokes/interleave and 1580 interleaves, resulting in 34760 spokes total, fat suppression, and scan time 3:55 minutes.

For ARKS, the same bSSFP scan parameters were used. However, at each TR, the current cardio-respiratory bin is identified and the optimal radial k-space sample to acquire is calculated. This is done by creating a Delaunay triangulation of the points already acquired and identifying the location of the Voronoi vertex associated with the largest circle (Fig. 1), which represents where data are missing in 3D radial k-space.

We compared the maximum Voronoi cell area across time and the scan efficiency, defined by time until the largest Voronoi area approached the area obtained by uniform sampling, for both approaches.

Results:

The average heart and respiratory rates of the 15 patients were 81.1±16.3 bpm (range 56.1-126.3 bpm) and 18.0±5.7 bpm (range 8.0-28.3 breaths/min), respectively. Each cardiorespiratory phase had on average 1158.7±79.9 spokes acquired (range 887 to 1366).

The distribution of 3D radial spokes improved via ARKS (Fig. 2, top row); the maximum area of Voronoi cells at the end of the scan was reduced by 48.1±15.1%. By efficiently targeting gaps in k-space, ARKS is more effective at acquiring data; ARKS scan efficiency (56.1±4.7%) was 1.9 times higher than GA-based spiral phyllotaxis (30.5±5.9%).

Conclusion:

Adaptive k-space sampling may be beneficial in 5D imaging approaches, particularly in pediatric patients who have higher heart and respiratory rates with physiologically more variation. We showed that 3D ARKS improved the distribution of k-space data and reached the same maximum Voronoi cell area in half the time as 3D spiral phyllotaxis. Future work will focus on implementing 3D ARKS in vivo and evaluating the effect of imaging trajectory on data quality.