A New Approach to Stable MRI Exams

Shigeru Kiryu, M.D., Ph.D.

Changing Requirements of MRI Exams

The global COVID-19 pandemic has greatly changed the environment surrounding Radiology. In last year’s RSNA2021, the theme of “Redefining Radiology” represented a defining frame for when we move past the pandemic. RSNA2021 highlighted workflow efficiency and patient-centeredness, while considering how to simultaneously maintain high-value imaging.
Prof. Kiryu (far right), with Department of Radiology staff from International University of Health and Welfare Narita Hospital, and Canon Medical representatives (suited in middle).
In order to provide high-value images, the key is how to avoid producing images with low diagnostic value. A typical example of a low diagnostic value image is an image with severe artifacts. In particular, when the artifacts cause difficulty in diagnosis, re-scanning may be required, impacting the overall efficiency of workflow. Certain MRI artifacts can be avoided by imaging conditions, and others are specific to patients due to motion. Even when simply referring to “motion”, there are various types of motion, such as breathing, heartbeat, blood flow, and accidental movements, including swallowing or coughing during the examination. In addition, the pattern of the artifacts varies depending on the characteristics of the motion. It is not easy to control the patient’s own motion, and it is possible to experience artifacts by accidental movement if the scan time becomes lengthened. Canon Medical Systems has commercialised new technology that suppresses these motion induced artifacts in order to improve workflow efficiency and provide stable exams, which I would like to introduce in the following paragraphs.

Counteracting Motion Artifacts

In principle, there are prospective and retrospective methods to suppress motion artifacts. The prospective method utilises physical sensors (including camera observation in recent years), a specific sequence called ‘navigator echoes’, or a part of the collected data to observe body movements, then the effects of movement will be eliminated or corrected by the observation. Respiratory and ECG synchronisation, Realtime Motion Correction (RMC) corresponds to this. This method can be used in combination with various sequences, and stable image quality can be obtained, but the scan time is often extended. Especially if the patient’s breathing and heartbeat are not stable, the scan time may be nearly twice as long as expected, or data collection may not proceed, and the examination may be interrupted. Alternatively, one of the well-known retrospective methods is to collect data radially in the k-space and correct this data during reconstruction. Canon Medical Systems provides this type of application known as JET. JET is used in combination with FSE2D (Fast Spin Echo 2D) sequence and is a very effective body motion correction technique. However, since the k-space sampling pattern is different from Cartesian sampling, which is often used in clinical practice, the contrast of images is likely to appear slightly different from that of general images and a specific streak artifact can occur.
Prof. Kiryu with Canon’s research level system, Vantage Centurian.
Iterative Motion Correction (IMC) application, a body motion correction technology which has recently been released by Canon Medical Systems, is equivalent to a retrospective method, but unlike JET, the data sampling pattern in the k-space is Cartesian. Therefore, this method addresses certain issues experienced with JET.
Prof. Kiryu reviewing images with a colleague.

Iterative Motion Correction (IMC) – an Alternative Solution

IMC is also used in combination with FSE2D sequences. In Canon’s latest software release (V8.0), the main target is Brain FLAIR (fluid-attenuated inversion recovery) and C-spine T2WI.
IMC correction can be applied to the motion of a rigid body, such as the head. The motion of the rigid body is simplified to estimate and correct the amount of translation and rotation. The process is shown in Figure 1. In general, FSE acquires data for image creation in each shot (Echo Train). In the case of IMC, however, the additional data for motion detection is collected in the same shot. IMC always uses a kind of parallel imaging reconstruction so that FSE data acquisition is under-sampled, and it contributes to scan time as short as that for a routine clinical scan. As shown in Figure 1, there are two processes for rigid body correction. Initially, shot rejection process uses motion detection data to eliminate large motion data. After that, rigid motion correction process uses the remaining data and performs to estimate both the amount of motion (motion estimation) and the images without motion (image estimation) by solving the unknown parameters.
Figure 1: Flowchart of Iterative Motion Correction reconstruction processing for rigid motion (left) and the comparison of the images with / without IMC (right)
Image estimation process uses the Conjugate Gradient SENSE (SENSitivity Encoding) method to estimate the expected final reconstructed image from under-sampling collected data1. Then, motion estimation process simulates k-space data with a certain amount of body motion by adding the motion parameters (translational and rotational movement amounts). The simulated data and the actual collected data are compared on an image domain to find the answer for the unknown parameters with minimum error. As the error cannot be minimised by this processing only once, both image estimation and motion estimation are repeated alternatively and iteratively. Finally, it is possible to obtain a reconstructed image that suppresses the influence of movement.

IMC corrects not only rigid motion, but also non-rigid motion. This correction process is shown in Figure 2.
Figure 2: Flowchart of Iterative Motion Correction reconstruction processing for non-rigid motion (left) and the comparison of the images with / without IMC (right).
As shown in the figure, rigid body motion correction is first performed using a 2D navigator echo. This is different from the previously explained method that executes rigid body correction and parallel imaging reconstruction alternatively, however, the target for motion to be corrected is the same translation and rotational movement. Shot rejection is then performed to maintain the correction accuracy by excluding the data with large changes. Then, non-rigid body correction process is executed. The non-rigid body correction utilises the process of creating synthetic data from under-sampling data in the k-space domain2. One line in the RO direction is considered in the k-space of the acquired data after rigid body correction. The acquired data are compared with the synthetic data created by estimating from the surrounding data. If the consistency is poor, the acquired data is likely to be affected by the patient’s motion, and it is replaced with the synthetic data. Thus, it is possible to prepare k-space data that suppresses the influence of motion as a whole, and finally, a reconstructed image in which the motion is suppressed can be obtained.

Example cases of Iterative Motion Correction

The effectiveness of IMC is shown in Figure 3. In this example, the volunteer intentionally moved during the scan. It is clear that the artifact of the motion is well suppressed by applying IMC.
Figure 3: Example of the images with / without Iterative Motion Correction reconstruction processing. A and C show the original images which have artifact induced by motion during the scan (yellow and red arrows). B shows the reconstructed image with IMC processing (only rigid motion correction) and D shows the non-rigid motion correction processed image. IMC effectively demonstrates reduced motion artifacts.
As mentioned earlier, the advantage of this IMC method is Cartesian collection, in which sampling patterns in k-space are conventionally used. As a result, it is possible to replace routine examinations without additional consideration.

The IMC can also be used in conjunction with Deep Learning Reconstruction, Advanced intelligent Clear-IQ Engine (AiCE). As shown in Figure 4, by using IMC in combination with AiCE, the stability improvement by SNR and motion suppression makes it possible to provide more valuable images, and it is highly anticipated that such new technologies improve the examination quality and future diagnosis in the post-COVID-19 Radiology world. //
Figure 4: Example of both IMC and AiCE applied image. As compared with the original image, the processed image (IMC with AiCE) has better image quality in terms of SNR and motion artifact suppression.
1 Pruessmann KP, et al., Magn Reson Med., 46:638-651, 2001
2 Huang Feng, et al., Magn Reson Med., 64(1):157-166, 2010
IUHW Narita Hospital.

Shigeru Kiryu, M.D., Ph.D.
Department of Radiology,
International University of Health and Welfare Narita Hospital, Japan.
Contact Us