3085

Towards motion-resolved interleaved radial ²³Na/¹H magnetic resonance imaging of the human heart at 7 Tesla

Jörn Huber¹, Laurent Ruck^2,3, Matthias Günther^1,4,5, Armin Nagel^2,6, and Simon Konstandin^1,4
¹Fraunhofer Institute for Digital Medicine MEVIS, Bremen, Germany, ²Institute of Radiology, University Hospital Erlangen, Erlangen, Germany, ³Friedrich-Alexander-Universität Erlangen-Nürnberg (FAU), Erlangen, Germany, Erlangen, Germany, ⁴mediri GmbH, Heidelberg, Germany, ⁵Faculty 1 (Physics/Electrical Engineering), University of Bremen, Bremen, Germany, ⁶Division of Medical Physics in Radiology, German Cancer Research Center (DKFZ), Heidelberg, Germany, Erlangen, Germany

Synopsis

Keywords: Image Reconstruction, Myocardium

Motion correction in interleaved 23Na/1H MRI of the human heart is important to improve the diagnostic reliability of reconstructed images and derived quantitative parameters. Therefore, this work demonstrates and compares the application of different reconstruction techniques to undersampled and motion-gated 23Na/1H MRI data at 7 T.

Introduction

²³Na concentration can be an important indicator for cell viability, making ²³Na MRI a promising technique for providing additional diagnostic information about dysfunctional myocardium^1,2. However, low signal-to-noise ratio (SNR) requires additional high-resolution ¹H images for segmentation of cardiac compartments and partial volume correction which prolongs scan time. This limitation can be overcome by an interleaved density-adapted radial ²³Na/¹H sequence scheme which enables the acquisition of cardiac ²³Na and ¹H images within a single scan. However, since typical scan times are in the range of minutes, interleaved ²³Na/¹H imaging still requires robust compensation of respiratory and cardiac motion prior to further analysis. Motion compensation can be accomplished by retrospective gating of acquired ²³Na/¹H projections. Reconstructed images from gated data however often suffer from low quality due to streaking artifacts arising from high undersampling factors. The BART toolbox³ offers a variety of algorithms, designed for artifact-free reconstruction of undersampled MRI data. To this aim, we evaluate different reconstruction techniques applied to motion-gated ²³Na/¹H MRI data to derive the optimal setup for future applications.

Methods

23Na and ¹H images were reconstructed from data which were acquired using a dual-nuclear interleaved sequence with a density-adapted 3D radial readout scheme^4,5. The interleaved sequence scheme allows to acquire four ¹H projections within a single ²³Na TR, thus being especially time efficient. Related sequence parameters are given in Fig. 1. In the given scenario, 15,000 ²³Na and 60,000 ¹H projections were acquired following a 3D golden angle scheme, which enables respiratory and cardiac gating. ¹H data were acquired using a sixteen channel receive coil, enabling parallel imaging reconstruction while ²³Na data were acquired using a single channel only. Acquired ²³Na/¹H data were reconstructed according to Figure 2a. ¹H projections were used for retrospective respiratory and cardiac gating. Two motion states were defined for each type of motion, resulting in a total number of four motion bins. PCA-based analysis of the k-space center signal of every fourth ¹H projection was used to define the motion states of the three preceding ¹H projections and the preceding ²³Na projection (cf. Fig. 2b). To evaluate optimal image reconstruction technique, different reconstruction mechanisms from the BART toolbox were applied to the gated ²³Na/¹H data. First, gridding reconstruction of motion-gated ²³Na/¹H images was applied in combination with pre-/post-density compensation, followed by a sum-of-squares coil combination (in case of multi-channel ¹H data). Furthermore, parallel imaging reconstruction techniques were tested on ¹H data due to the multi-channel readout. Therefore, all non-gated ¹H projections were first used to reconstruct 3D low-resolution images (40x40x40 voxels) of individual coils which were used to estimate coil sensitivity profiles using different algorithms (cf. Fig. 2c). The different calibration commands are given in Fig. 1b and the calibration technique, yielding the best image quality, was further used to assess the effect of different regularization terms during the image reconstruction process.

Results

Estimated sensitivity profiles and their respective effects on ¹H image reconstruction are shown in Fig. 3. Individual reconstructed ¹H images using ESPIRiT reconstruction⁶ with different L2 regularization weights are compared to standard gridding reconstructions in Figure 4. Figure 5 finally compares reconstructed motion-gated ²³Na/¹H images using gridding only (²³Na)/gridding and ESPIRiT reconstructions (¹H).

Discussion

The calibration of ¹H coil sensitivity maps showed to have a direct impact on the resulting image quality. Here, the ESPIRiT calibration showed a reduction of artifacts (cf. arrows in Fig. 3) when compared to the other methods and was thus applied during the following parallel imaging reconstructions. The standard gridding reconstruction with pre-density compensation resulted in sharp ¹H images with residual streaking artifacts, appearing in a noise-like fashion (cf. Fig. 4). Post-density compensation appears to reduce noise-like streaking artifacts at the cost of reduced sharpness. Strong L2 regularization resulted in strong image blurring and is thus not recommended. In comparison, the ESPIRiT reconstruction further enhances the signal in the heart region (cf. Fig. 4 arrows) but also shows a loss in detail structures, especially at higher L2 regularization factors. In general, it appears that L2 regularization has no benefits for the image reconstruction applied in this work. The same trend shows when comparing gridding and ESPIRiT reconstructions for ¹H data in Fig. 5. In case of ²³Na data, the gridding routine with post-density compensation seems to be beneficial for reduction of noise-like artifacts. However, it should be investigated whether the same can be achieved using additional filter operations (e.g., Hanning) in case of gridding with pre-density compensation. It should also be mentioned that the presented data suffers from severe B1 inhomogeneity which results in signal loss in structures distant to the transmit coils. Parallel transmit techniques could therefore be beneficial to homogenize the signal intensity.

Conclusion

This work evaluated reconstruction algorithms from the BART toolbox in combination with interleaved and motion-gated ¹H/²³Na MRI data. A combination of ESPIRiT reconstruction of ¹H data and gridding reconstruction with post-density compensation of ²³Na data can be recommended to reduce the noise in final images. However, future work must investigate, whether similar or better results can be achieved using gridding alone with additional filter operations or other reconstruction frameworks such as XD-GRASP⁷.

Acknowledgements

This project was funded by the Deutsche Forschungsgemeinschaft (DFG) under project number 449552397.

References

1. Joern J W Sandstede, Hanns Hillenbrand, Meinrad Beer et al., Time course of ²³Na signal intensity after myocardial infarction in humans. Magn Reson Med . 2004 Sep;52(3):545-51

2. Johanna Lott, Tanja Platt, Sebastian C Niesporek et al., Corrections of myocardial tissue sodium concentration measurements in human cardiac ²³Na MRI at 7 Tesla. Magn Reson Med . 2019 Jul;82(1):159-173

3. Uecker et al., The BART Toolbox for Computational Magnetic Resonance Imaging, Proc Intl Soc Magn Reson in Med, 2016

4. Armin M Nagel, Frederik B Laun, Marc-André Weber et al., Sodium MRI using a density-adapted 3D radial acquisition technique, Magn Reson Med . 2009 Dec;62(6):1565-73

5. Ruck et al., Interleaved ²³Na/1H MRI of the human heart at 7 Tesla, Proc Intl Soc Magn Reson in Med, 2022

6. Martin Uecker, Peng Lai, Mark J Murphy et al., ESPIRiT--an eigenvalue approach to autocalibrating parallel MRI: where SENSE meets GRAPPA, Magn Reson Med . 2014 Mar;71(3):990-1001

7. Li Feng, Leon Axel, Hersh Chandarana et al., XD-GRASP: Golden-angle radial MRI with reconstruction of extra motion-state dimensions using compressed sensing, Magn Reson Med . 2016 Feb;75(2):775-88

Figures

Figure 1: a.) Sequence parameters used for the in-vivo scan in this work; b.) BART commands for sensitivity estimation and respective effects; and c.) BART commands for image reconstruction and respective effects.

Figure 2: a.) Image reconstruction pipeline of 1H/23Na data; b.) Binning algorithm used to extract respiratory and cardiac signals/bins; and c.) Estimation of coil sensitivities as applied in this work.

Figure 3: Estimated coil sensitivity profiles using different configurations available in the BART toolbox. Note that two maps were estimated using the ESPIRiT command (lower row). Images were reconstructed using the BART pics command with a small L2 regularization parameter.

Figure 4: Evaluation of different reconstruction techniques using 1H data from the first motion bin. Arrows indicate areas in the heart with SNR variations using the different reconstruction techniques.

Figure 5: Reconstruction of a total number of four motion bins (two respiratory, two cardiac) using different reconstruction methods. The upper half shows reconstructions from 1H data while the lower half shows reconstructions of 23Na data. A different colormap was applied for visualization purposes.

Proc. Intl. Soc. Mag. Reson. Med. 31 (2023)

3085

DOI: https://doi.org/10.58530/2023/3085