Synopsis

Keywords: Heart, Motion Correction

In this work, we develop, validate, and apply a novel reconstruction framework for intra-bin correction and inter-bin compensation of respiratory motion in cardiac and respiratory motion-resolved free-running whole heart 5D MRI. We demonstrate respiratory resolved images with reduced artifact and without compressing the underlying physiological motion providing a means to increase the acquired resolution, accelerate the acquisition, or investigate respiratory driven changes in cardiac output and blood flow, potentially leading to new MRI driven biomarkers for cardiovascular disease

Background

Cardiac and respiratory motion-resolved whole-heart 5D imaging has been shown to improve acquisition efficiency^1,2 while also providing a means of evaluating respiratory driven changes in cardiac output and blood flow³. However, two limitations of respiratory-resolved imaging exist which are related to user-defined parameters. First, the number of reconstructed respiratory phases presents a trade-off between the level of undersampling, which may lead to residual artifact, and the amount of respiratory motion within each reconstructed phase (intra-bin), which may lead to image blur. Second, a weighting parameter in the reconstruction must be chosen such that undersampling artifact is reduced without introducing blur due to the remaining motion between each reconstructed phase (inter-bin). Overall, these parameters may be subject-specific and may limit our achievable resolution and acceleration factor.
In this work, we develop and validate strategies for intra- and inter-bin respiratory motion compensation to improve the quality of whole-heart 5D MRI in terms of sharpness, residual artifact level, and fidelity of the underlying respiratory motion^1,2,4. We validate our approach using a comprehensive numerical simulation and further compare reconstructions with and without the proposed motion compensation strategies in a cohort of 16 patients. With the goal of an improved reconstruction that may enable increased resolution or a reduction in scan time, we test the hypotheses that the inclusion of non-rigid deformation fields in the image reconstruction will enable larger regularization weights to reduce artifact without incurring blur and that the combined approach for intra- and inter-bin respiratory motion compensation will result in improved image quality for whole-heart 5D MRI.

Methods

For our proposed reconstruction framework (Fig. 1), we adapted a previously established method (fNAV) for estimating respiratory motion in 3D radial data⁵ to correct the motion within the bins (intra-bin) of our reconstructions. Additionally, we estimated nonrigid deformation fields between adjacent frames (inter-bin)⁶. Finally, we combined the estimated intra-bin correction and inter-bin compensation with cardiac and respiratory self-gating to bin the data and reconstruct 5D images². To validate this framework, a numerical simulation of free-running 3D radial data was developed^5,7,8. Simulated data (N=50), each with variable respiratory motion amplitude and heart rate, were generated to match in vivo sequence parameters. To demonstrate the feasibility of our approach in vivo, 16 patients with congenital heart disease were included in this IRB approved study. Examinations were performed without sedation, during free breathing, on a 1.5T clinical MRI system (MAGNETOM Sola, Siemens Healthcare, Erlangen, Germany) after administration of ferumoxytol. A slab-selective spoiled gradient echo free-running 3D radial research sequence⁴ was used and resulted in uninterrupted acquisitions of six minutes^1,6. Relevant parameters were RF excitation angle: 15°, resolution: (1.15 mm)³, FOV: (220 mm)³, TE/TR: 1.53/2.84 ms, readout bandwidth: 1002 Hz/pixel. To test their impact on reconstructions with and without motion compensation, respiratory regularization weights were varied between 0 and 0.25 and the resulting images from both simulated and in vivo data were compared. Reconstruction error (root mean squared difference) over a region of interest containing the lung-liver interface was calculated using the ground truth as reference for simulation and a reconstruction containing all the cardiac bins and without compressed sensing for in vivo data. Compression of respiratory motion was quantified by measuring the displacement of the liver relative to the reference images. Sharpness of the lung-liver interface was quantified as the slope of a sigmoid function fitted to the interface⁹. Using the optimum regularization weight in terms of these three metrics, 5D images from all simulated and in vivo datasets were reconstructed and the error over a region containing the heart was quantified in simulation data where ground truth is known, and the image quality of in vivo data was assessed using an artificial-intelligence based algorithm trained to grade 3D radial images of the heart¹⁰. Statistical significance was measured using a paired t-test.

Results

Reconstruction error, motion compression, and sharpness measurements from simulated (Fig. 2a-c) and in vivo (Fig. 2d-e) data demonstrate a consistent statistically significant improvement across the majority of the tested regularization weights when using motion compensation. In vivo image reconstructions (Fig. 3) corroborate the quantitative results from Fig. 2 and display the improved image quality provided by the proposed method, despite increasing regularization weights. For both simulated (Fig 4a) and in vivo (Fig. 4b) data, metrics for image quality are significantly higher using motion compensation. Finally, an animated depiction of 5D images (Fig. 5) using the optimized parameters shows a clear improvement in image quality, particularly during the mid and end-inspiration phases.

Discussion and Conclusions

A novel framework for respiratory motion compensation was developed, validated, and applied to whole heart 5D MRI in 16 patients. Quantitative assessment of image quality yielded statistically significant improvements when using the proposed framework while in vivo findings corroborated the results from numerical simulations. Overall, this work not only highlights potential sources of error when reconstructing respiratory resolved images without compensation but provides a robust solution that enables future studies that aim to increase the achievable resolution, decrease acquisition time, and investigate respiratory driven changes in cardiac output and blood flow, potentially leading to new MRI driven biomarkers for cardiovascular disease.

Acknowledgements

MS is the PI on the Swiss National Science Foundation grants 320030_173129 and 201292 that funded part of this research. CWR is the PI on Swiss National Science Foundation Grant PZ00P3_202140 that funded part of this research.