0486

Self-gated ²³Na-MRI of human lung with separate reconstruction of two respiratory states at 7T

Tanja Platt¹, Reiner Umathum¹, Armin M. Nagel^1,2, Peter Bachert¹, Mark E. Ladd¹, Mark O. Wielpütz³, Hans-Ulrich Kauczor³, and Nicolas G. R. Behl¹

¹Medical Physics in Radiology, German Cancer Research Center (DKFZ), Heidelberg, Germany, ²Institute of Radiology, University Hospital Erlangen, Erlangen, Germany, ³Diagnostic and Interventional Radiology, University Hospital Heidelberg, Heidelberg, Germany

Synopsis

²³Na-MRI at 7T can give valuable insights into lung physiology because ²³Na signals are linked to sodium-potassium-ATPase and to fluid balance. However, low NMR sensitivity and low in-vivo concentrations of Na⁺ ions in-vivo limit the achievable spatial resolution and prolong acquisition times. In this study a retrospective self-gated reconstruction was used to reduce motion artifacts in free-breathing ²³Na-MRI of human lung. The presented method allows for the reconstruction of ²³Na images which correspond to two respiratory motion states for free-breathing volunteers and patients. 3D Dictionary-Learning Compressed-Sensing reconstruction was shown to markedly reduce image noise.

Introduction

MRI of human lung is challenging due to low spin density, susceptibility artifacts, and physiological motion¹. Compared to proton MRI, sodium MRI (²³Na-MRI) is less sensitive to susceptibility effects due to the lower Larmor frequency. Moreover, ²³Na-MRI yields valuable information on lung physiology because ²³Na signals are linked to sodium-potassium-ATPase and to fluid balance. Hence, ²³Na-MRI has been proposed for investigating lung tumor viability and potentially treatment response², as well as for investigating fluid distribution in edematous lung^1,3,4. However, low NMR sensitivity and low in-vivo concentrations of Na⁺ ions in-vivo limit the achievable spatial resolution.

Progress in ²³Na-MRI techniques like ultra-high field, dedicated hardware, UTE sequences, and compressed-sensing reconstruction algorithms expand the quality of ²³Na-MRI of human lung^5,6. However, low ²³Na concentrations prolong acquisition times beyond human breath-hold capabilities.

In this study we use a golden angle sampling scheme with retrospective self-gated reconstruction⁷ to reduce motion artifacts in free-breathing ²³Na-MRI of human lung. This enables the reconstruction of two motion states, providing insights into variations of ²³Na distribution during respiration. Resulting under-sampled data subsets are reconstructed using a 3D Dictionary-Learning Compressed-Sensing reconstruction (3D-DLCS) to reduce image noise⁸. In-vivo results with t_acq=30min 20s are presented.

Methods

Measurements were performed on a 7T whole-body MR scanner⁹ using a 4-port elliptically-shaped birdcage¹⁰ (length: 35cm). Phantom and in-vivo ²³Na images (healthy volunteer, female, 27y/o) were acquired with a 3D density-adapted radial sampling scheme¹¹ and a golden angle distribution of projections¹².

Parameters: t_p=2ms, T_E/T_R=1.05ms/20ms, nominal resolution=(4mm)³, FA=44°, t_acq=30min 20s, N=18200 projections, reconstructed FOV=(400mm)³, fivefold short-term averaging.

After averaging, time-resolved intensity at k-space center was determined for every projection (S_{center,p=1…N}, Δt_center=5∙TR=100ms). For comparison (Fig.1a-c), an external signal from a respiratory belt positioned on the chest was obtained (Δt_external=20ms).

A 1D Fourier transform (FT) along k-space center points was performed for phantom and in-vivo MRI data, as well as along data points from the respiratory belt to evaluate corresponding frequency spectra. The relevant frequency domain for respiration (0.1Hz-0.6Hz) is compared for all three datasets (Fig.1d-f). The frequency distribution around 0.27Hz in the in-vivo k-space center spectrum can be attributed to respiratory motion, as it does not appear in the phantom spectrum but is visible in the respiratory belt data.

The frequency band from ca. 0.2Hz to ca. 0.35Hz is used to obtain the respiratory signal S_resp(t) from the data acquired for ²³Na-MRI of human lung (Fig.2). Each projection is assigned to either the inspiratory or the expiratory data subset (D_insp if S_resp(t)<0, D_exp if S_resp(t)>0).

Under-sampled data subsets with numbers of projection N_inspiration=9063 and N_expiration=9137, as well as the full dataset (N=18200) are reconstructed with a Nonuniform Fast Fourier Transform (NUFFT)¹³ and with 3D-DLCS (Fig.3). Spatial intensity variation due to respiration is analyzed for the region of liver and diaphragm (Fig.4).

Results

The time-resolved intensities in k-space center for in-vivo and phantom data are shown in (Fig.1a,b) together with the corresponding spectra (Fig.1d,e). The comparison of phantom and in-vivo spectra suggests that the distribution between ca. 0.2Hz and ca. 0.35Hz can be assigned to respiratory motion. This is confirmed by the spectrum from the respiratory belt (Fig.1f).

Respiratory signal S_resp(t), which is extracted from the same data acquired for ²³Na image reconstruction, corresponds very well to signal recorded by the respiratory belt (Fig.2d,e). During expiration (blue), signal from respiratory belt is low, while k-space center signal is high because more ²³Na containing tissue is located in the sensitive part of the coil and vice versa for inspiration (red).

Motion-separated lung images reconstructed with NUFFT show a considerable degradation of image quality compared to the NUFFT image reconstructed with all projections. The 3D-DLCS reconstruction, however, is less prone to higher under-sampling as it occurs when different motion states are separated. Generally, lung images reconstructed with 3D-DLCS exhibit markedly reduced image noise compared to NUFFT images. The effect of respiratory motion is illustrated exemplarily for the region of liver and diaphragm (Fig.4).

Discussion & Conclusion

Retrospective self-gated reconstruction of ²³Na datasets of human lung is feasible for the setup and sequence parameters used here, making a respiratory sensor obsolete. This allows for the reconstruction of ²³Na images which correspond to different motion states for free-breathing volunteers and patients. In the present work the number of respiratory states is limited to two due to the low SNR of ²³Na-MRI. Iterative reconstruction, like 3D-DLCS, was shown to reduce image noise and could enable the reconstruction of additional motion states. Reduced motion blurring and image noise may permit the detection of smaller lesions, which could be of interest when applying ²³Na-MRI to studies of lung pathologies.

Acknowledgements

This work was funded in part by the Helmholtz Alliance ICEMED - Imaging and Curing Environmental Metabolic Diseases, through the Initiative and Networking Fund of the Helmholtz Association.

This study was supported by grants from the Bundesministerium für Bildung und Forschung (BMBF) to the German Center for Lung Research (DZL) (82DZL00401 and 82DZL004A1).

References

1) Kauczor, H.U. & Kreitner, K.F. Eur Radiol (1999) 9: 1755-64

2) Henzler, T. et al. Rofo (2012) 184: 340-4

3) Kundel, H.L. et al. Magn Reson Med (1988) 6: 381-9

4) Lancaster, L. et al. Magn Reson Med (1991) 19: 96-104

5) Madelin, G. & Regatte, R.R. J Magn Reson Imaging (2013) 38: 511–29

6) Lustig, M. et al., Magn Reson Med (2007) 6: 1182-95

7) Feng, L. et al. Magn Reson Med (2014) 72: 707-17

8) Behl, N.G.R. et al. Magn Reson Med (2016) 75: 1605-16

9) MAGNETOM 7T, Siemens Healthcare, Erlangen, Germany

10) Platt, T. et al. Proc Intl Soc Mag Reson Med (2016) 24: 3977

11) Nagel, A.M. et al. Magn Reson Med (2009) 62: 1565-73

12) Chan, R.W. et al. Magn Reson Med (2009) 61: 354-63

13) Fessler, J.A. et al. Trans Signal Process (2003) 2: 560-74

Figures

Fig.1: Temporal and spectral signals for motion detection. Top: Time-resolved signal intensity at k-space center of in-vivo (a) and phantom ²³Na-MRI (b), Δt_center=5∙TR=100ms and signal recorded by respiratory belt (c), Δt_external=20ms. Bottom: Frequency spectra of temporal signals derived from 1D FT showing relevant respiratory frequency domain (0.1Hz-0.6Hz). The in-vivo k-space center spectrum (d) displays a frequency distribution around 0.27Hz that is not present in the phantom spectrum (e) but corresponds to the spectrum of the respiratory belt (f). A peak arising from the acquisition scheme (d+e, marked with *) does not coincide with the respiratory frequency distribution (green box).

Fig.2: Separation of motion states. Each projection is sorted into one of two motion states via base-line correction of the in-vivo k-space center spectrum. a) Lower and upper respiration frequency limits (f_lower, f_upper, green box) are determined. b) In-vivo k-space center spectrum is low-pass filtered twice: S(f>f_upper)=0, S(f>f_lower)=0. c) Temporal signals (S_upper(t), S_lower(t)) are obtained from low-pass filtered spectra via 1D FT. d) Respiratory signal is determined by subtraction: S_upper(t)-S_lower(t). Each projection is assigned to one data subset: D_inspiration for S_resp(t)<0 (red), D_expiration for S_resp(t)>0 (blue). e) Comparative external signal from a respiratory belt showing equivalent but inverse signal variation.

Fig.3: ²³Na-MRI of human lung at B₀=7T for full dataset and for separated motion states. The acquisition time of the single underlying dataset was ca. 30min. (a,d) Full dataset (N=18200 projections) and data subsets (b,e) for inspiration (N_inspiration=9063) and (c,f) expiration (N_expiration=9137) are reconstructed with NUFFT (top) and with 3D-DLCS reconstruction algorithm (bottom). Separation of motion states leads to considerable degradation of image quality for NUFFT reconstruction, while for 3D-DLCS image noise is comparable to that of the full dataset. In general, ²³Na images reconstructed with 3D-DLCS exhibit markedly reduced image noise compared to the images reconstructed with NUFFT.

Fig.4: Respiratory motion is illustrated for liver and diaphragm region (a) in a line plot and (b) in a difference image along the white dashed lines and within the white boxes in Fig.3e and 3f. During expiration (a, blue line) liver is located further in head direction, resulting in higher intensities at z=28-32 compared to the inspiration case (a, red). The image in (b) shows the relative deviation between the inspiratory image and the expiratory image, revealing respiratory motion.

Proc. Intl. Soc. Mag. Reson. Med. 25 (2017)

0486