Prospective motion correction (PMC) using marker tracking has great potential to compensate for patient motion during MRI experiments. However, one challenge with marker tracking during MRI experiments is the coupling of the marker to the organ of interest. For brain imaging, markers are usually affixed to the forehead, goggles or to the teeth of the patient. Although the fixture to the teeth gives the most reliable tracking data of the skull, it is uncomfortable and costly. The adhesion of markers onto the forehead of the patient is currently the most convenient option from the patients’ perspective. [1]

A limitation to this marker fixation approach is that skin on the forehead does not always remain coupled to the motion of the skull. [2] Facial gestures such as squinting can lead to spurious position data. One example of a falsely motion corrected MR image is depicted in Figure 1. The corresponding position data are depicted in Figure 2, showing the relative translations and angular positions between two markers affixed to the forehead. Variations in the relative positions of the two markers (orange curve) imply that skin motion has happened. This false position data causes severe motion artifacts observed in the MR image.

This work presents a solution to make prospective motion correction significantly more robust against squinting motion by tracking of multiple markers and the introduction of a “virtual marker”.

In order to obtain matching information about the skull motion of the subject, multiple markers from a Moiré Phase Tracking (MPT) System (Metria Innovation, USA) were affixed to the face of the subjects. By applying three or more markers it is possible to calculate additional “virtual markers” from the correlation of the positions of the individual markers. Thus a “virtual marker” is created from at least three marker positions. The virtual markers give position data in 6 degrees of freedom (DOF) but are literally independent on the individual marker orientations.

The following steps are performed, to calculate the virtual marker:

1. Record the simultaneous position of 3 individual MPT markers.

2. Determine in-plane vectors between the marker positions and the corresponding surface normal.

3. Determine orthonormal basis, based on one of the in-plane vectors and the surface normal.

4. Determine the rotation matrix from change in the orthonormal basis to the next frame.

In the following experiments, two markers are positioned on the forehead approximately 1 cm above each eyebrow, and another positioned on the cheek. An additional reference marker was attached to a mouthpiece worn by the subject. The mouthpiece was assumed perfectly coupled to the skull. For Experiment One the patient was told to move the head side-to-side and to suppress facial expressions. For Experiment Two, the patient was encouraged to squint and grimace during the head motion.

The rotation log of Experiment One depicted in Figure 3 shows that the blue, cyan and green curves which correspond to individual MPT markers are in agreement with the black reference curve. The red curve illustrating the virtual marker is also in agreement with the black reference curve, but exhibits greater noise. This is further quantified in Table 1, which depicts the mean squared error (MSE) compared with the reference marker. All MSE for Experiment One are of the same order of magnitude, indicating that the MPT markers and the virtual marker follow the reference marker equally accurately.

Figure 4 shows the robustness of the virtual marker against facial expressions during Experiment Two. As the curves of the MPT markers (green, blue, cyan) deviate significantly compared to the black reference curve, the virtual marker curve (red), remained in adequate agreement. For squinting motion the MSE is an order of magnitude higher for two of the facial markers in comparison to the virtual marker. This suggests that the virtual marker can be used to stabilize the tracking data. In the optimal case the virtual marker will only be used if the MPT markers are affected by facial expressions.

For marker based tracking methods multi marker tracking poses a solution to stability problems caused by skin motion. While using a mouthpiece for marker fixation remains the most reliable method, the virtual marker approach provides a substantially more robust approach compared to a single marker attached to the face. The presented work illustrates a robust solution for marker based head tracking during MRI examinations, and owing to the cost-effectiveness of this approach, may find relevance in clinical settings. One limitation however is that at least 3 markers need to be continuously visible.