PT - JOURNAL ARTICLE AU - James Goddard AU - Mark Mandelkern AU - Megan McClintick TI - PET Head-Motion Correction by facial feature tracking DP - 2019 May 01 TA - Journal of Nuclear Medicine PG - 1355--1355 VI - 60 IP - supplement 1 4099 - http://jnm.snmjournals.org/content/60/supplement_1/1355.short 4100 - http://jnm.snmjournals.org/content/60/supplement_1/1355.full SO - J Nucl Med2019 May 01; 60 AB - 1355Objectives: Head motion is a significant problem in PET brain scanning, especially for patients in altered mental states. Blurring due to head motion causes misestimation of physiological parameters, such as binding potentials for neuroligands. Head restraints are poorly tolerated and don't meet the requirements of modern scanners, roughly 1 mm of motion. Optical devices for tracking use markers that are difficult to attach rigidly. Data-based methods require long scan subintervals to minimize noise, and tracers that bind widely in the brain without rapid changes in distribution to obtain clear features for realignment. Our method using optical measurement of facial features for fast and accurate pose determination. Methods: The stereo optical system views the patient’s head in the PET position and performs 2D face detection. A light source provides indirect illumination. A 3D measurement of head features is made for each camera frame at 6 per second. These are compared to that for an early reference frame, to determine relative pose as a function of time. Pose error is not cumulative and is relatively insensitive to facial movements (e.g., mouth, eyes, facial contortions) independent of the overall head motion. RMS error of 3D point alignment is less than 0.5 mm in manikin studies. An offline calibration step is performed to determine the camera-to-PET 3D reference frame. An example of the pose alignment for a human study is given, where an unaligned pair of 3D views is shown along with the aligned view. We performed human scans at a Siemens mCT with the high-affinity dopamine D2 neuroligand F-18 fallypride. For an accurate measurement of binding potential, one must scan for roughly three hours. Our protocol consists of a CT scan followed by two 80-minute PET blocks separated by a break. Point sources were applied for orientation and to assess motion. F-18 fallypride predominantly binds to the striatum, with order-of-magnitude smaller binding to the cerebral cortex, cerebellum, and other subcortical structures. Striatal binding potentials are the measures of interest, and motion degrades accuracy for these small structures. Motion correction was performed during reconstruction as described in Reference 1, Method C. We derived 16 10-minute images from sets of 1-minute sinograms using OP-OSEM (3 iterations, 21 subsets). Attenuation maps are created for each 1-minute interval. Scatter and random corrections are applied. Processing took approximately 21 minutes per 10-minute frame on a Windows laptop with an i7-8550U CPU @1.80 GHz and 16 GB of RAM. Comparisons of images with and without correction were made and preliminary binding potentials were computed. Results: RMS error of 3D point alignment is approximately 1 mm in human studies. Small structures with high fallypride uptake are qualitatively sharper after motion correction. BP values are generally increased for small structures. Conclusions: Motion tracking of the head using optical measurement of facial features is a convenient and accurate method for motion-correction of PET images, yielding high-quality images and accurate physiological measurements. Acknowledgements: We appreciate the contributions of Mike Casey, Inki Hong and Judd Jones at Siemens.