Improving Respiratory Phase-resolved 3D Body Imaging Using Iterative Motion Correction and Average (MoCoAve)
Xiaoming Bi1, Jianing Pang2, Wensha Yang2, Matthias Fenchel3, Zixin Deng2, Yuhua Chen4, Richard Tuli2, Debiao Li2, Gerhard Laub1, and Zhaoyang Fan2

1Siemens Healthcare, Los Angeles, CA, United States, 2Cedars-Sinai Medical Center, Los Angeles, CA, United States, 3Siemens Healthcare GmbH, Erlangen, Germany, 4University of Pennsylvania, Philadelphia, PA, United States

Synopsis

4D (respiratory phase-resolved 3D) MRI has been increasingly used for the planning of radiotherapy and minimally invasive surgery. Recently developed self-gating methods showed great potential in 4D MRI by providing high imaging efficiency and isotropic spatial resolution. However, images of individual phases may suffer from decreased SNR and increased streaking artifact since only a subset of data were used for reconstruction. A motion correction and average (MoCoAve) framework was developed in this work to address such limitations. Preliminary results from patients showed that the proposed method can significantly improve SNR and image quality without compromising motion information.

Purpose

4D (respiratory phase-resolved 3D) MRI has been increasingly used for the treatment simulation and adaption for radiotherapy, as well as planning of minimally invasive surgery. Recently developed 4D self-gating methods [1-3] overcomes spatiotemporal limitations of conventional 2D real-time and 3D navigator-gated methods. Using motion information directly derived from self-gating signal, k-space data are retrospectively grouped into multiple bins representing different breathing phases. Since each bin only contains a subset of the acquired data, reconstructions from individual bin could suffer from low SNR and severe streaking artifacts. Reducing the number of bins alleviates the undersampling effect within each bin, yet will introduce blurring as the motion may not be adequately resolved. In this work, we hypothesize that the quality of 4D images can be improved without compromising motion information by averaging the undersampled bin images after motion correction (MoCoAve).

Methods

Two prototype sequences [2-3] were used for acquiring free-breathing 4D MR images using stack-of-stars and koosh-ball radial trajectories, respectively. As shown in Fig. 1, acquired imaging data were evenly grouped into 10 respiratory bins using self-gating signals extracted either from k-space centers (stack-of-stars) or a self-gating line (koosh-ball) in the superior-inferior direction. A 3D image was then reconstructed from each bin using the respective subset of data. The MoCoAve process was subsequently performed on the undersampled 10-phase image series. For example, if image set 1 is chosen as the reference bin, then the forward and inverse transform between image sets 2-9 and image 1 are computed using a symmetric diffeomorphic model [4]. Then, the MoCoAve image of each phase is produced by first aligning all phases to this phase using the respective combinations of transforms, and then averaging the warped images to enhance image quality. Five patients with confirmed tumor (3 pancreatic, 1 liver, 1 lung) were scanned on 3T clinical scanners (MAGNETOM Verio or Biograph mMR, Siemens Healthcare, Germany). Seven 4D MR scans (4 koosh-ball) were acquired with the following parameters: FLASH readout; FA = 10°; FOV = 400 mm. Stack-of-stars: 104 partitions, 6/8 partial Fourier, 1504 projections, (1.98 mm)^3 voxel size. Koosh-ball: 73005 projections, 256 reconstructed partitions with (1.56 mm)^3 resolution. MoCoAve processing was performed offline using MATLAB. SNR and motion trajectory of tumors were assessed from each respiratory phase using images directly reconstructed from individual bins, as well as corresponding images after MoCoAve. SNR was defined as signal intensity in the liver divided by the standard deviation of background air signal. For each patient, ROIs for signal measurement were matched between different image sets. For motion assessment, tumor volumes were drawn on the end expiration phase of the 4D MR images, and then mapped to the other phases using a B-spline based deformable registration in VelocityAITM (Varian, Palo Alto, CA). The coordinates of the contours’ center of mass were then extracted for tumor motion trajectory evaluation. Measurement results of SNR and tumor motion before and after MoCoAve were compared using paired t-test.

Results

4D MR images were successfully acquired from all subjects. Imaging time was 9.0 and 6.5 minutes, respectively, using stack-of-stars and koosh-ball trajectories. Fig. 2 shows representative stack-of-stars and koosh-ball images from two patients. Note visually improved SNR, reduced streaking artifacts and preserved sharpness using the proposed MoCoAve method. Quantitative analysis showed significantly improved SNR (mean±SD without and with MoCoAve: stack-of-stars: 7.5±2.1 vs 21.3±5.9, p < 0.01; koosh-ball: 28.9±13.1 vs 43.2±19.1, p<0.01). Fig. 3 shows example 4D MoCoAve images at two different phases with resolved respiratory motion. Figure 4 illustrates the motion trajectory of a lung cancer nodule measured from 10 bins without and with MoCoAve. The motion trajectory of tumors measured from 70 volumetric image pairs showed excellent agreement between images reconstructed without and with MoCoAve. Correlation coefficients were: 0.94±0.10, 0.88±0.12, and 0.74±0.16 in the SI, AP and LR directions.

Discussion

With the proposed MoCoAve method, the entire k-space data contribute to the reconstruction of each respiratory phase. Compared with previous methods that only use a subset of data, it has the benefits of improved SNR and reduced streaking artifacts for each respiratory phase. Preliminary studies show measured motion trajectory of tumor is not compromised. Such a method can be used for improving the quality of 4D MR images, or accelerating 4D MRI by using fewer projections since SNR loss in individual bins can be alleviated by the MoCoAve process.

Conclusion

A motion corrected averaging method was developed for improving the quality of 4D MRI while preserving the motion information.

Acknowledgements

No acknowledgement found.

References

[1] Buerger C et al, IEEE TMI 2013, p805.

[2] Grimm R et al, Medical Image Analysis 2015, p110.

[3] Deng Z et al, MRM 2015.

[4] Avants BB et al, Med. Image Anal. 2008, p26

Figures

Figure 1: schematic diagram of the proposed MoCoAve method for 4D MRI. K-space data were sorted into 10 bins based on respiratory self-gating signal and reconstructed into 10 image sets. Motion correction was performed toward a target image (set 1 in this example) prior to averaging of all warped images. Such process was repeated to generate MoCoAve images for all respiratory phases.

Figure 2: Representative stack-of-stars (top row) and koosh-ball images from two patients. Images are directly reconstructed from all acquired k-space lines without binning (left column), one respiratory bin with 10% of data (middle column), and corresponding MoCoAve images (right column). Note improved delineation of tissue boundaries (arrows) and vessel delineation (ovals) using MoCoAve.

Figure 3: Representative 4D MoCoAve images at inspiratory (left column) and expiratory phases (right column). Images were acquired using stack-of-stars trajectory (top row) and koosh-ball trajectory (bottom row), respectively, from two patients.

Figure 4: Relative motion trajectory measured from a lung cancer nodule (red circles in the PET and MR images). Motions measured from original image sets (dashed lines) and MoCoAve image sets were closely correlated in all three directions.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
0617