Synopsis

Due to the long scan times of seconds or even minutes, MRI is susceptible to subject motion. Such motion can lead to ghosting, blurring and other image artifacts and can result in non diagnostic images or false quantitative results in clinical and scientific studies. Faster imaging is a convenient method to avoid or reduce motion artifacts but has limitations in terms of resolution, and image quality. Motion correction, therefore, is a research field with a long history but only few methods entered clinical routine. A number of approaches have the potential for broader application.

• Navigators are the largest group of methods. 1D or 2D navigators have been proposed a long time ago. More recently, full volume navigators have been introduced. They acquire low resolution 3D volumes or a number of orthogonal slices with fast imaging such as spiral or EPI. An alternative approach with less disturbance of water magnetization are fat navigators, wich selectively acquire a 3D fat volume and are particularly effective in head imaging. These navigators require significant scan time and are thus most effectively integrated in 3D methods with recovery periods, such as MPRAGE or 3D TSE. They also allow pose tracking with relatively slow update rates and are thus best suited for correction of slow motion and drifts.
  
• Self-navigation is a subgroup of such navigator methods. Motion, or better pose information, is encoded and extracted from the image data itself. The most prominent example is PROPELLER (or BLADE), which acquires rectangular blocks through the center of in k-space that are rotated and thus allow low resolution reconstructions to correct for between-shot motion. These methods oversample parts of the data with some redundancy.
  
• Tracking with external sensors allows very fast pose detection and is independent of the MR measurement, therefore not disturbing the magnetization steady state. Tracking methods include optical methods, electromagnetic tracking, inertial (MEMS) sensors, field probes and others. Fast, accurate and sequence-independent motion information, however, requires additional hardware and calibration of the coordinate systems.
• Fully data driven motion correction estimates parameters of a motion model to optimize image quality according to a defined metric. Such methods do not require any additional hardware or measurement time. Even model-free approaches using deep learning networks have been proposed. While prospective correction methods can only account for rigid body motion (i.e. rotation and translation), data driven correction can account for more complex motion patterns.

Acknowledgements

References