PET/MR motion correction

Overview

The following figure outlines the system overview:

Acquisition

For a short scan time of 1-2min, a reliable motion model can be generated, freeing time for additional diagnostic MR sequences. A 3D Cartesian time-independent variable-density subsampling with possible ESPReSSo mask and a self-navigation approach is applied. With the help of external sensor signals (ECG, camera signal, ...) and a sensor fusion approach, a respiratory surrogate signal for the complete scan time can be derived.

Reconstruction

After extracting the self-navigated MR signal and sensor signals, a sensor mapping is trained and all signals are fused, yielding a respiratory surrogate for the complete scan time. Based on the respiratory and cardiac surrogate, the acquired samples are gated, resulting in a 6D subsampled k-space (3D + t_resp + t_card + N_Rx,channels). A joint Compressed Sensing reconstruction and motion estimation procedure returns a motion-resolved MR image and a motion model.

The motion model together with an attenuation map derived from the reconstructed MR images enables to perform the PET motion correction.

All reconstruction steps are carried out in Gadgetron.

Download

Acquisition

The precompiled acquisition sequence, termed CS_Retro (Siemens, VB20P) can be downloaded:

https://github.com/thomaskuestner/CS_MoCo_LAB/acquisition/CS-Retro

or

https://github.com/thomaskuestner/4DMRImaging/acquisition

Reconstruction

A motion-resolved MR image can be retrieved via

• Matlab: https://github.com/thomaskuestner/4DMRImaging/reconstruction
• Gadgetron: https://github.com/thomaskuestner/CS_LAB/reconstruction/gadgetron/CS_LAB_Gadget
• Data emitter and injector functors: https://github.com/thomaskuestner/CS_LAB/reconstruction/gadgetron/CS_LAB_Gadget/src/GADGET_TOOLS
• Please follow the guideline for setting up Gadgetron on a Siemens Biograph mMR.

Please cite one of the papers, if you use it in a scientific publication.

References