MR scientists and physicians interested in lung MRI and the assessment of regional pulmonary ventilation.

Due to its patient friendly free-breathing acquisition without use of any contrast agent, Fourier Decomposition¹ (FD) is a lung function proton imaging technique with a high clinical potential. FD was validated in animal and human studies^2,3 and methods for quantification were introduced.^4-6 Nevertheless, FD is inherently limited by its 2D acquisition. The scan of a whole lung requires approximately 10 minutes and it can be assumed that thoracic breathing can cause through plane motion leading to artifacts.

Self-navigated MRI sequences allow the reconstruction of 3D images in different respiratory states without the requirement for breath-holds.⁷ Similar to density changes measurements between expiration and inspiration on chest CT, these images can be used to quantify ventilation. The purpose of this study is to demonstrate the feasibility of 3D ventilation maps and a direct comparison with 2D FD.

Six healthy volunteers were enrolled in this study. The protocol contained coronal FD scans without a slice gap covering the whole lung and an additional 3D volume scan of the whole lung. All acquisitions were performed on 1.5T scanner during free breathing.

For FD, 200 images were acquired for each slice using a spoiled gradient echo sequence (FOV=50x50 cm², matrix size=196x196, slice thickness=15 mm, TE=0.94 ms, TR=3 ms, flip angle=5° and GRAPPA=2) over a period of 65 s at a temporal resolution of 322 ms. After image registration (ANTS)⁸ image analysis¹ and FV quantification^5,6 was conducted.

For 3D calculation, 3496 spokes were acquired using a stack-of-stars gradient echo sequence with a golden angle increment (FOV=50x50 cm², matrix size=196x196x36, slice thickness=5 mm, TE=0.92 ms, TR=3 ms and flip angle=5°) over a period of 6.3 min. The DC signal was used for sorting the spokes into six uniform datasets according to the respiratory phase⁹. Then, the 3D images were reformatted to a slice thickness of 15 mm. The image at end-expiration (I_exp) was registered to an image at end-inspiration (I_insp). FV was calculated voxel-wise with the registered images according to: FV = (I_exp-I_insp)/I_exp.⁵

Large vessels were excluded by manual segmentation. Mean FV values were calculated for FD (FV_2D) and for the proposed method (FV_3D) and compared as a function of slice position. Using all voxel values mean values of the whole lung were calculated as well.

Figure 1 shows FV_2D and FV_3D maps of a healthy volunteer for anterior to posterior slices. Both methods display high vessel/parenchyma sharpness and a good visual agreement. Nevertheless the white arrows indicate regions on the FV_3D maps, which display low FV values not present on the FV_2D map.

The evaluation of FV_2D as a function of slice location shows a high variability of FV values and no evident slice dependent behavior (see Figure 2a). Contrary, for FV_3D all volunteers show the same pattern: Higher values towards the posterior and anterior slices (see Figure 2b).

Analyzing the difference between FV_2D and FV_3D as a function of slice location shows that the highest deviations are found for the anterior slices (see Figure 3). Additionally, there is an increase from the more stable values of the middle slices towards the posterior slice locations. Interestingly, the volunteers with the highest deviations are also the subjects with the highest FV values (compare to Figure 2a).

For mean FV values of the whole lung both methods showed a very good agreement (R² = 0.88, Figure 4).

This study shows that self-navigation can be used to calculate 3D FV maps, which show a good agreement with 2D FD for the middle slices and larger deviations for the anterior and posterior slices. The latter results can by the explained by through plane motion, which affects the anterior and posterior slices to a greater extent compared to the middle slice. Consistently, the highest deviations occurred for subjects with the highest tidal volume. Through plane motion and FV variability due to different tidal volumes between the slice scans is a likely reason for a missing evident slice location dependent behavior for 2D FD, which is clearly visible on FV_3D. Also, 3D FV values were systematically higher compared to the 2D FV values, which may be due to the fact that the complex 3D lung parenchymal motion is more adequately captured using the 3D technique. The implementation of compressed sensing and parallel imaging algorithms could be used to further improve the image quality of the 3D method.In combination with the fast total acquisition time this method is a very attractive alternative to 2D FD imaging.