Yajing Zhang1, Zhen Jiang2, Weiping Liu1, Feng Huang1, Ming Yang1, Allan Jin1, and Ping Yang1
1Philips Healthcare, Suzhou, China, People's Republic of, 22nd affiliated Hospital of Soochow University, Suzhou, China, People's Republic of
Synopsis
Quantification of dynamic contrast enhanced MRI (DCE-MRI) is
often hindered by motion during imaging. Multiple sources of motion require for
a non-rigid 3D registration to align dynamic images. This study provided a
rapid 3D non-rigid registration tool for DCE liver registration by optimizing
the scheme of image matching. The results show that the dynamic images were
well aligned in terms of whole liver area and the portal vein area. Meanwhile,
the intensity plot demonstrated better representation from the registered
images. Computation time of registration was about one minute for the entire
scan, making it possible for clinical routine analysis.Target audience
Radiologist and clinical researchers who are interested in
liver perfusion quantification and post-processing.
Purpose
We aim at providing a practical dynamic contrast enhanced MRI
(DCE-MRI) liver registration tool, with high speed for clinical research.
Quantitative analysis on liver function has become increasingly important for DCE-MRI
and the diagnosis of lesions. However, the quantification was often misconducted due
to motion-induced mismatch of images among different dynamic phases. Image
registration approaches make it possible to analyze the haemodynamic change in
liver on a voxel or region of interest (ROI) level. Since multiple sources of
motion contributed to the mismatch, a non-rigid registration is required.
Previous methods usually take a few minutes to register one dynamic scan images,
which is unacceptable for clinical practice. This study proposes a rapid 3D
non-rigid registration tool for DCE liver registration by optimizing the scheme
of image matching. To evaluate the registration performance, computation time
and ROI analysis have been shown.
Methods
Clinical routine
DCE-MRI was conducted on a 61 year-old male subject with lung cancer suffering diarrhea,
in the 2nd affiliated Hospital of Soochow University. The imaging
was performed on a 3 T Ingenia system (Philips Healthcare, the Netherlands), with
acquisition of 64 slices per scan with 12 dynamic scans (parameters: TR = 3.51ms,
flip angle = 10°, FOV = 400*400 mm2, matrix size: 480*480, slice
thickness = 3 mm). Contrast agent was given after acquiring three scans as
reference phase. Image registration was applied on the image series to align
all images to the 1st time point image. The registration was based
on freeform deformation [1] and mutual information to enable
application on images with different contrasts [2, 3], e.g. DCE
images. We reduced the number of sampled points for fast calculation of
transformation, so that the computation time was greatly reduced. Registration
was performed using an HP Z620 Workstation. For evaluation of registration
performance, three-dimensional whole liver boundary and the main portion of
portal vein were delineated manually on the 1st time point image and
the mean intensities of the two ROIs on each time point were calculated.
Results and discussion
No abnormality of any abdomen organ was detected
on the subject by radiologist. Selected time point images from preparation
phase and 3 perfusion phases were shown in Fig.1 to evaluate the registration
performance. The dynamic images were well registered to the pre-contrast image,
with respect to both the whole liver boundary (ROI in cyan) and the portal vein
(ROI in pink). Red arrows in Fig.1 highlighted the severe mis-match of original
scans. Fig.2 demonstrated the intensity plot of the ROIs. The mean intensity
curve after registration (Fig.2 - right) matched the previous report from literature
[4], while quantification from original images (Fig.2 - left) was
contaminated by mis-match error. Computation performance of registration was
shown in Table 1. The registration only took around 5 seconds for aligning one scan
image, with memory used of 472 megabytes. The total time for registration was
within 1 minute, much faster than the traditional non-rigid 3D registration
method [5]. This acceleration was achieved by the optimized portion
of sampled voxels for calculation of the transformation.
Conclusion
The optimized 3D liver
registration tool rapidly aligns DCE liver images with local contrast changes.
The registration improves the ROI-based quantification of liver and portal
vein. The
good alignment of liver and other tissue organs will be helpful for voxel-based
analysis of lesion and k-trans analysis. Because of the rapid computation with
satisfied registration performance, this tool is promising to be applied in DCE
image analysis.
Acknowledgements
No acknowledgement found.References
[1] Lee,
S., et al. Image Metamorphosis with Scattered Feature Constraints. IEEE Trans.
Vis. Comput. Graphics, 1996; 2(4):
337-354.
[2] Maes,
F., et al. Multimodality Image Registration by Maximization of Mutual
Information. IEEE Trans. Med. Imaging, 1997 ; 16(2) : 187-198.
[3] Tohlfing, T., et al. Volume-Preserving
Nonrigid Registration of MR Breast Images Using Free-Form Deformation
With an Incompressibility ConstraintIEEE
Trans. Med. Imaging, 2003; 22(6): 730-741
[4] Do,
R.K.G, et al. Dynamic Contrast-Enhanced MR Imaging of the Liver : Current
Status and Future Directions. Mag. Reson. Imag. Clin. North America, 2009;
339-349
[5] Maes,
F., et al. Comparative evaluation of multiresolution optimized strategies for
multimodality image registration by maximization of mutual information. Med.
Imag. Analysis, 1999 ; 3(4) : 373-386