Synopsis

Keywords: Digestive, Motion Correction

Diffusion-weighted MRI is increasingly used for detection and characterization of Crohn’s disease. However, unavoidable respiratory motion and bowel motility reduces accuracy and precision of quantitative parameter fitting, which hinders clinical applicability DW-MRI. We use a 3D slice-to-volume registration approach that sequentially tracks rigid motion parameters for each slice and regularises the parameters with a Kalman filter in the order of acquisition of each slice. We assess the quality of images and estimated parameter maps and the precision of IVIM parameters in the areas of disease using the proposed motion correction technique, and compare them with results from the uncorrected data.

Introduction

Diffusion-weighted MRI signal is increasingly used in assessment of Crohn’s disease (CD) owing to improvements in MRI hardware¹. In comparison to the single exponential signal decay model, bi-exponential intravoxel incoherent motion model (IVIM) can be used from multi-b-value DW-MRI, which provides a more accurate slow diffusion decay parameter (D) and additional fast diffusion coefficient fraction (f) parameter².
The model is given by:
$$S(b)=S_{0}\left(f e^{-b D^{*}}+(1-f) e^{-b D}\right)$$
where D* is fast diffusion parameter, S0 is non-diffusion dependent signal, and b is b-values vector.
IVIM parameters can be used for detection of lesions and characterization of inflammation and fibrosis³. Nevertheless, due to unavoidable respiratory motion and motility of bowel, the accuracy and precision of the quantitative parameter fitting is reduced, which hinders clinical applicability. Respiratory motion can be compensated using respiratory triggering, however this method does not correct for bowel motility. Previous methods used non-rigid registration and model based registration for motion correction in abdominal DW-MRI^4-6. However limitations persist due to the ill-posedness of the non-rigid registration. Here we use a 3D slice-to-volume registration (SVR) approach that sequentially tracks rigid motion parameters for each slice and regularises the parameters with a Kalman filter along the order of slice acquisition^7-8. We assess the quality of images and parameter maps and the precision of IVIM parameters in the areas of CD using the proposed motion correction technique, and compare them with results from the uncorrected data.

Methods

DW-MR images from 11 pediatric Crohn's disease patients were acquired clinically at 3T (MAGNETOM Prisma, Siemens) using a free-breathing single-shot EPI: TR/TE=8100/67;matrix=156x120;FOV=340x260mm;slice-thickness=5mm;b-values=0,20,50,100,200,400,600,800s/mm2;6 gradient directions;acquisition-time= 6.5min. Images were retrospectively reviewed according to the IRB protocol. Post-contrast T1 and DW-MRI were assessed by a trained radiologist and CD regions were segmented.
Six parameters of 3D rigid motion for each slice were first estimated using 3D SVR. The slice level motion tracking based on Kalman filtering was then used to regularise motion parameters for each slice in the order of acquisition.
First, b=0s/mm2 image with least amount of motion was used as the initial reference volume. To prevent motion estimates to skew towards non specific regions of interest, the volume was cropped around the Crohn’s disease area. Next, motion parameters for each slice in the b=0 volumes were sequentially estimated by registering the reference volume to each slice.Then, updated average B0 image was reconstructed. This averaged B0 image is then used as reference to register the next set of b values, and the process is repeated for all b-values.
Estimated motion parameters were applied to each slice for motion correction using a scattered point cloud interpolation scheme, which uses a regular reconstruction grid to map a collection of Gaussian kernel weighted voxels from motion space. The entire motion estimation and reconstruction algorithm was written in C++ with ITK and ANIMA^9,10, and is available on github.com/quin-med-harvard-edu/dSVRK, and as a docker image hub.docker.com/r/quinlab/dSVRK. Next, IVIM model was fitted to uncorrected and motion corrected data for estimation of D and f parameters using DIPY library¹¹.
To assess improvement in precision of estimated parameters after motion correction, bootstrap analysis was used. Random removal of gradient directions in the original data was performed to generate subsampled data points at each bootstrap iteration. In total 61 iterations were performed. IVIM model was repeatedly fit to each bootstrap subsampled signal. Evaluation consisted of the coefficient of variation estimation of the D and f parameters over multiple bootstrap iterations.

Results

Figure 1 shows the quality of IVIM model parameter maps before and after motion correction. The image quality of the parameter maps improved significantly after motion correction in all subjects.
Figure 2 shows the quality of b=50,400,800s/mm2 images before and after motion correction in 3 orthogonal planes. The areas with motion, the structural details in the images and the consistency of the anatomy were all improved after motion correction.
Figure 3 shows a line of voxels plotted over time in the order of slice acquisition for all volumes of an example subject. If there was no motion, straight lines would appear across time. Motion corrected data showed more consistent lines over time compared to uncorrected data.
Figure 4 shows the results from the assessment of precision of IVIM parameter estimates via bootstrap analysis on the entire dataset. Motion corrected data exhibits significant reduction in CoV of parameters, which is driven by significant reduction in standard deviation of parameter values after alignment of multiple b-value images.
Figure 5 shows improved anatomical consistency between T1 images and DW-MRI after motion correction.

Conclusions and Discussion

The proposed 3D slice level motion tracking method could correct for the motility and respiratory related motion of Crohn’s disease regions in DW-MRI. Motion correction significantly improved precision of IVIM parameters by reducing the CoV% of the estimated IVIM parameters. The motion corrected images and the resultant parameter maps all showed improved image quality, improving the consistency of structures and reducing the misalignments in both spatial and temporal dimensions. Despite the nonrigid nature of motion, a local 3D rigid SVR method that tracked the 3D rigid motion of each slice was effective in correcting motion in DW-MRI of Crohn’s disease.

Acknowledgements

This work was supported in part by NIH grants R01 EB019483, R01 NS121657, R01 DK125561, R21 DK123569, R21 EB02962, and a pilot grant (PP-1905-34002) from the National Multiple Sclerosis Society.