Image Reconstruction System for Compressed Sensing Retrospective Motion Correction for the Application in Clinical Practice
Martin Schwartz1,2, Thomas Küstner1,2, Christian Würslin1, Petros Martirosian1, Nina F. Schwenzer3, Fritz Schick1, Bin Yang2, and Holger Schmidt3

1Section on Experimental Radiology, Department of Radiology, University of Tuebingen, Tuebingen, Germany, 2Institute of Signal Processing and System Theory, University of Stuttgart, Stuttgart, Germany, 3Diagnostic and Interventional Radiology, Department of Radiology, University of Tuebingen, Tuebingen, Germany

Synopsis

Respiratory motion-free images are important in MRI of the human thorax and abdomen. A significant factor is the reconstruction of these images for the application in clinical practice. The objective of the presented work is an integration of an approved motion correction algorithm into the clinical environment to overcome limitations of offline reconstructed images by the utilization of external workstations. Therefore, a reconstruction pipeline based on the open-source framework Gadgetron with new modules for the integration of a motion correction algorithm is demonstrated.

Purpose

One challenging task in MRI of the human thorax and abdomen is to get respiratory motion-free images to prevent image degradation. For this reason, many different algorithms for the detection and correction of motion-induced artifacts were investigated over the past years 1,2,3. Most of them were restricted to an offline reconstruction on an external workstation and thus restricted to research purposes limiting the use in a clinical setup. Furthermore, this offline reconstruction prevents the utilization of vendor-specific correction methods, e.g., gradient distortion correction, and forces additional effort for the operator after image acquisition, e.g., reading and processing of raw data format. For the utilized algorithm 4,5, we have to acquire a 4D motion model as fast and accurate as possible, hence the method relies on a subsampled acquisition which demands later a Compressed Sensing reconstruction. Due to this, a high performance of the reconstruction system is required to ensure a fast image reconstruction. Instead of the inflexible implementation of the reconstruction algorithm in the image calculation environment (ICE) of the vendor system and a more comfortable handling by starting the acquisition, viewing the results on the host system and saving them in standard DICOM format, an external data processing with communication to the vendor system is desired. To overcome these limitations and to provide a more flexible implementation, the objective of this work is a smooth integration of an approved motion correction 4,5 into the clinical environment with the reconstruction on an external workstation via Gadgetron 6.

Methods

Acquisition: Data of a free-breathing subject is acquired continuously based on a standard 3D spoiled gradient echo sequence with random subsampling in phase-encoding direction and a self-navigation approach 4. Reconstruction: The proposed reconstruction system is based on the open-source framework Gadgetron 6, which enables the reconstruction on powerful external workstations. Based on this setup, new functional blocks, called Gadgets, were implemented into Gadgetron for the motion correction system, as well as for the transmitting and receiving of the raw k-space data in the vendor (Siemens, VB20P) reconstruction pipeline. The motion correction system is depicted in Fig. 1. First, the navigator signal is extracted from the incoming data stream and the respiratory gates can be determined by different methods like equidistant gating or k-means. The subsampled k-space is reordered and populated based on the respiratory gates 4. The missing k-space data is calculated based on a Compressed Sensing reconstruction. We implemented a 4D FOCUSS algorithm 4,7 with the optional extension of the ESPReSSo constraint 8. Instead of reconstructing the image and bypass the vendor reconstruction pipeline, the k-space data is directly passed back to the vendor reconstruction system providing the utilization of the vendor implemented data correction methods (Fig. 3). Therefore, the 4D k-space data is split into individual raw data acquisitions and the incomplete data header information is corrected (Fig. 2). For registration of the different motion states, the vendor corrected image data is sent again to the workstation by a new implemented image data emitter (Fig. 3), which in addition can be easily utilized for the integration of other reconstruction procedures. The image registration itself is implemented in a new module providing the access to the registration toolbox elastix 9,10. The utilization of this toolbox provides a comfortable and flexible parametrization with well-known parameter files. The proposed system, which integrates the Gadgetron-based motion correction reconstruction pipeline into the vendor reconstruction pipeline (ICE), was implemented on a 3 T Biograph mMR (Siemens Healthcare, Erlangen, Germany) and is depicted in Fig. 3. The reconstruction setup can be parameterized by the user interface on the scanner. The implemented Gadgetron-based motion reconstruction pipeline containing the new modules will be made available under BSD license at: https://sites.google.com/site/kspaceastronauts/.

Results

An exemplary coronal acquisition, which was acquired with a total scan time of 191 s, is depicted in Fig. 3 (right corner). Imaging parameters: matrix size: 256x256x72, FoV = 500 x 500 mm², TE = 1.71 ms, TR = 3.4 ms, BW = 1030 Hz/px, navigator period = 200 ms. The image was reconstructed by the 4D FOCUSS algorithm without additional ESPReSSo constraint. The four respiration states were registered by means of a non-rigid multilevel cubic B-spline transformation of the elastix toolbox.

Conclusion

The proposed reconstruction system provides integration of a motion correction reconstruction into clinical setting, which enables a clinical feasible motion correction workflow resulting in improved image quality by allowing for the utilization of the vendor-provided correction methods. The full integration enables analysis of larger studies with better comparability to other motion correction algorithms.

Acknowledgements

No acknowledgement found.

References

[1]: Würslin et al., Journal of Nuclear Medicine 2013:54(3); [2]: King et al., Medical Image Analysis 2012:16(1); [3]: Grimm et al., Medical Image Analysis 2015:19(1); [4]: Küstner et al., Proc. ISMRM 2014; [5]: Küstner et al., IEEE Proc. ICASSP 2015; [6]: Hansen et al., MRM 2013:69(6); [7]: Gorodnitsky et al., IEEE Trans. Signal Process. 1997:45(3); [8]: Küstner et al., IEEE Proc. ISBI 2014; [9]: Klein et al., IEEE Trans. Medical Imaging 2010:29(1); [10]: Shamonin et al., Front. Neuroinformatics:7(50)

Figures

Fig. 1: Motion correction system.

Fig. 2: k-space to raw data conversion with header correction.

Fig. 3: Image reconstruction system for the integration of a motion correction procedure into the clinical environment with highlighted new Functors and Gadgets.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
1845