Abstract
3320
Introduction: Patient motion has limited the Positron Emission Tomography (PET) image quality and quantification accuracy. For high resolution PET imaging of a rigid human brain, both the position and orientation of the head must be precisely estimated in real time. Recent motion tracking (MT) technologies utilize infrared optical systems to construct the 3D profile of an object along the camera's line-of-sight (e.g., human subject's face) and track motion using pattern recognition. However, such optical systems are complex, expensive, prone to occlusion (especially in small-diameter brain PET scanners), and cannot track non-rigid motion. We propose our cost-effective, non-optical, and wearable electromagnetic MT system without line-of-sight restrictions which with careful design and engineering has overcome the electric and magnetic interferences and achieved precise 6 degree-of-freedom (DOF) motion capture at sampling frequency exceeding 100 Hz.
Methods: The schematic diagram of the MT setup is shown in Fig. 1a where a source emits an electromagnetic dipole field and is positioned in close proximity to the sensors using an articulating arm support. A wearable and compact 6 DOF sensing system (wired or wireless) is attached to the head using a form of headwear (e.g., headband, headliner, etc) and detects changes of the electromagnetic flux density. The MT readout digitizes the signal from an array of sensors and calculates the precise translational and rotational motion by mitigating environmental distortions. Figure 1b shows the custom-designed phantom where two fillable Locus Coeruleus (LC) regions have been added to a conventional 3D Huffman brain phantom. Each LC pocket is 2 mm in diameter and spaced 3 mm from the midline. The phantom was filled with 1.85 mCi 18F-FDG (77 uCi/mL in the LC pockets, 11.2 uCi/mL in the rest of the phantom), and it was initially placed at the center of the scanner's field of view (FOV) and further translated to different positions using a motorized xyz stage to simulate motion (Fig. 1c). The high resolution Prism-PET brain scanner, which can resolve hot spots with diameters as small as 1.35 mm, was utilized for our MT experiments. Figure 1d shows the phantom with the sensors taped at the back of the phantom and the source placed inside the gantry. Three MT experiments were performed to test the accuracy of our electromagnetic MT system given the gantry's eddy current distortion which is of significant concern. The Prism-PET scanner was electrically turned on in the first and second experiments, and then turned off in the third experiment. The phantom imaging and motion correction were performed in the second experiment where real-time motion data and PET coincidence events were acquired.
Results: The translation errors of the three MT experiments were < 1 mm at each position by averaging the results of an array of sensors, shown in Fig. 1f. The average translation error for experiments 1, 2, and 3 were 0.24, 0.34 and 0.28 mm, respectively. The images without motion correction (top-row images in Fig. 1g) were obtained by directly combining the acquired PET data in all positions and consequently severe blurring was observed with LC regions completely washed out. Next, both the ground truth (GT) and MT data of phantom translations were used to perform motion correction in image space (middle-row and bottom-row images in Fig. 1g). Given that the MT accuracy was substantially better than the scanner's spatial resolution, both motion-corrected images performed identically and successfully in resolving the right and left LC nuclei with high signal-to-noise ratio.
Conclusions: Our proposed cost-effective electromagnetic MT system mounted on the high resolution Prism-PET brain scanner achieved submillimeter MT accuracy with mitigated environmental distortions. Motion correction in image space using real-time MT data enabled high resolution visualization of the bilateral LC regions with high contrast.