Patient motion during an MR scan is still a problem in the medical field. It results in artefacts and reduces image quality which can lead to wrong patient outcomes. A patient moves due to the long scanning times, discomfort, and restlessness. Motion correction techniques are developed to prevent or compensate for the movements. Prospective motion correction is one of the available options. The motion is corrected after the image is made. Optical motion tracking systems can be used to track patient motion. To determine the pose of a patient, tracking markers are used. Research is done in in-bore and out-of-bore camera systems. Here one or more cameras are used to track markers on a subject. In in-bore systems, the camera is placed more closely to the marker, whereas in out-of-bore systems, the cameras experience less distortion from the MR. Other options for optical tracking systems are the use of lasers or optical fibres. Almost all studies use tracking markers, whereas some investigated markerless tracking. Most research was not suitable for clinical implementation. In research, less is known about the effect of marker design on the accuracy of a camera system.
This thesis investigates how marker design can help improve motion detection systems, specifically Philips’s VitalEye camera. This is done by developing a model that represents the VitalEye camera using rigid transformations, a Pinhole camera model and an image formation model called rasterization. The pinhole camera model uses the properties of similar triangles to establish an image point. The focal length, the number of pixels, and the pixel width and height can transform the image points into digital images. A cost function and the Jacobian are established. Using the Newton-Raphson method, the transformation matrix can be determined. The complete VitalEye model is validated. The marker placement by the model and in the image were in good agreement, and the resulting transformation vectors had only minor errors. Eight different shapes are investigated with different sizes and orientations.
It was found that accuracy increases with an increase in marker size and a decrease in distance. Additionally, two-dimensional (2D) shapes showed lower accuracy than prism shapes. The best 2D shape was the 100x100 [mm] square (median:upper adjacent:maximum, z: 0.45:1.65:2.60 [mm]; 𝛼:0.75:2.56:3.99 [∘]). The prisms showed that thicker shapes improve the rotation accuracy around the x- and y-axis.
Changing the orientations did not improve the accuracy. From the prisms, the 100x100x40 [mm] cube (median:upper adjacent:maximum, z: 0.41:1.35:2.10 [mm]; 𝛼:0.08:0.28:0.43 [∘]) and the 100x80x40 [mm] hexagonal prism (median:upper adjacent:maximum, z: 0.46:1.68:2.23 [mm]; 𝛼:0.09:0.30:0.48[∘]).