4480

Navigator-less Spiral SToRM for Free breathing and Ungated Cardiac CINE MRI

Abdul Haseeb Ahmed¹, Ruixi Zhou², Yang Yang², Michael Salerno², and Mathews Jacob¹

¹Electrical Engineering, University of Iowa, Iowa city, IA, United States, ²Medicine and Biomedical Engineering, University of Virginia, Charlottesville, VA, United States

Synopsis

This study introduces an iterative kernel low-rank algorithm to recover images in a free breathing and ungated cardiac MRI dataset. The approach relies on the manifold structure of dynamic data to recover it from highly undersampled measurements. The data is acquired using variable density spiral acquisition. An iterative kernel low-rank algorithm is introduced to estimate the manifold structure of the images, or equivalently the manifold Laplacian matrix, from central k-space regions. Unlike previous manifold regularization implementations, the iterative algorithm, coupled with the non-Cartesian acquisitions, eliminates the need for dedicated navigators to estimate the manifold Laplacian, thus improving sampling efficiency.The iterative kernel low-rank algorithm facilitates the extension of manifold regularization to navigatorless spiral acquisitions, thus improving sampling efficiency. This algorithm provides improved reconstruction compared to the state of the art methods.

Background

Self-gating cardiac MRI strategies were introduced to overcome the challenges of current breath-held ECG-gated cardiac MRI acquisition techniques. Traditional self-gating strategies rely on explicit binning of data into distinct cardiac and respiratory phases, followed by image reconstruction. The recent smoothness regularization on manifold (SToRM) approach [1], which performs implicit binning of data, provides improved reconstruction of images in phases with less data. A challenge with SToRM is the need for explicit k-space navigators to estimate the manifold structure, which reduces the sampling efficiency.

Methods

In this work, we introduce an iterative SToRM algorithm, which does not require explicit k-space navigators. We acquire the data using variable density spiral trajectories; the manifold structure is estimated from the central k-space regions with higher sampling density. Since the central k-space regions are still under sampled, we propose to use an iterative kernel low-rank method to estimate the manifold structure. The estimated manifold structure is used to recover the images from highly under sampled multi-coil spiral data in a computationally efficient manner as shown in a Figure 1. Spiral data was collected on a 3T Siemens scanner using a golden-angle rotated gradient echo spiral pulse sequence with one interleaf per excitation. Sequence parameters included: TR 7.8ms, slice thickness 8mm, resolution 1.25mm² and total acquisition time of 16 seconds per slice. The navigator-less SToRM technique was compared against a low-rank method, 1st iteration SToRM where the manifold structure is directly estimated from multi coil SENSE reconstructed central k-space regions, and to SPARCS method [2], which is an explicit binning scheme.

Results

The data was acquired from four subjects with variable breathing patterns and heart rates. Figure 2 shows the performance of the proposed method, compared to a low rank method and the first iteration of SToRM using simulated cardiac data. Figure 3 and Figure 4 show the performance of the proposed method, compared to a low rank method and the first iteration of SToRM using free breathing and ungated data, which all aim to recover the entire image time series from the measurements. Figure 5 shows a comparison between the proposed technique and SPARCS scheme, which is a self-gated strategy that aims to recover a single time series of images covering the cardiac cycle. The proposed method demonstrates superior performance as compared to the existing schemes in terms of spatial and temporal fidelity. To show the performance of our proposed method, we have compared our scheme against a low-rank method, 1st iteration SToRM (Figures 2,3,4) and SPARCS that relies on binning and motion compensation (Figure 5).

Discussion

In this study, navigator-less spiral SToRM is introduced to overcome the limitation of SToRM algorithm. This approach recovers free breathing and ungated cardiac data without the requirements of explicit binning and navigator data. Results show that the less blurred images are achieved while maintaining the temporal fidelity. With the sharp endocardium and epicardium edges, we can use these images to measure the functional parameters of the heart. At the same time, with good temporal fidelity, we can also assess individual cardiac cycles which could be particularly useful in patients with irregular heart rhythms.

Conclusion

Navigator less SToRM is proposed to recover the free-breathing and ungated cardiac images from variable density spiral acquisition. Both spatial images and temporal profiles show significantly reduces spatial and temporal blurring with the proposed algorithm, in comparison with the existing techniques.

Acknowledgements

No acknowledgement found.

References

1. Poddar, Sunrita, and Mathews Jacob. "Dynamic MRI using smoothness regularization on manifolds (SToRM)." IEEE transactions on medical imaging 35.4 (2016): 1106-1115.

2. Zhou R, Yang Y, Matthew R, Salerno M. “Cardiac and Respiratory Self-Gated Motion-Corrected Free-Breathing Spiral Cine Imaging, Proceedings of the 26th ISMRM 2018

Figures

Figure 1: Outline of the Spiral-STORM method. The free breathing and ungated data is acquired using a golden angle spiral acquisition scheme. Two-step strategy, first recovering a low-resolution dataset, followed by high-resolution recovery, to realize a computationally efficient algorithm. We estimate the Laplacian matrix using an iterative kernel low-rank based method from the central k-space regions, illustrated with a square. Laplacian matrix specifies the connectivity of the points of the manifold, with larger weights between similar frames in the dataset. Once the Laplacian is available, the high-resolution is estimated from k-space samples by solving [8] method gives non-smooth temporal profile.

Figure 2: Simulated short axis cardiac CINE dataset using low rank method (b1-b3),SNR=18.68db, 1st Iteration of SToRM method (c1-c3), SNR=18.96db and the proposed method (d1-d3), SNR=30.44db. Each scheme is compared against the original dataset(a1-a3). This dynamic dataset is retrospectively undersampled using golden angle spiral sampling pattern. Three cardiac phases are picked from each reconstruction method which correspond to end of systolic, mid phase, and end of diastolic as shown by red, yellow, green lines in the time profile (a4). We observe that the proposed method provides reconstructions with lower spatial and temporal blurring, compared to other algorithms.

Figure 3: Volunteer one, performance comparison between Low Rank, Ist iteration SToRM [2], and the proposed method. Three frames at different phases along cardiac cycle are shown along with the temporal profile along the yellow line. The low rank and 1st iteration SToRM resulted in blurred images as compared to our method, which is providing sharper endocardial and epicardial boundaries. Since 1st iteration SToRM method [2] uses multi coil SENSE images for weight matrix estimation, it results in a smooth temporal profile without over regularization. The temporal profile of our method is comparable to the 1st iteration SToRM.

Figure 4: Volunteer two, performance comparison between Low Rank, Ist iteration SToRM [2], and the proposed method. This dataset has more respiratory motion displacement as compared to the volunteer one. The low rank and 1st iteration SToRM resulted in blurred images as compared to our method. The temporal profile of our method is comparable to the 1st iteration SToRM, whereas the low rank method gives non-smooth temporal profile. We conclude that our method results in cine images reconstruction with higher spatial and temporal fidelity.

Figure 5: Performance comparison between the L+S SPARCS method and the proposed method. Three columns show images at different phases of the cardiac cycle. The fourth column depicts the temporal profile across the middle of the left ventricular (yellow line). The proposed method yields sharper myocardium boundaries as compared to the L+S SPARCS method. The temporal profile of our method is comparable to the L+S method, as both methods show smooth transition between the end of systole and the end of diastole. Our method preserves spatial resolution without compromising temporal resolution and does not require explicit binning of data.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)

4480