Abstract
370
Objectives: In both clinical (e.g., pediatric) and research settings it is desirable to reduce the amount of administered radiopharmaceutical to reduce ionizing radiation exposure. However, PET image quality and diagnostic value is reduced if motion occurs during acquisition, which is manifested by blurring and inaccurate quantitation. With low dose imaging, motion management is a key issue due to the need for extended acquisition times. Proposed solutions for motion management in PET brain imaging include image registration techniques (which suffer from low temporal resolution, and are incompatible with low dose protocols), MR-data driven approaches (which require a simultaneous PET/MR scanner, and also interfere with availability of MR for simultaneous diagnostic imaging), and external tracking systems (which require integration of an external tracking system). In this work, we apply an optical camera tracking system to demonstrate the capability of low dose PET brain imaging with motion management using list mode reconstruction.
Methods: One subject was scanned on a GE SIGNA PET/MR system (GE Healthcare, Waukesha, WI). Motion tracking was performed with an optical camera system (HobbitView Inc., San Jose, CA) attached to the 8-channel head coil. A curved marker with unique labels was placed on the subject’s forehead to capture rigid motion (6 degrees of freedom: translations and rotations). A time of flight (TOF) PET scan was acquired in the head for approximately ten minutes (632 seconds). TOF-OSEM image reconstruction was performed using GE Healthcare’s list mode reconstruction software in MATLAB (R2018b, Mathworks, Natick, MA) with standard clinical settings. List mode PET data was retrospectively sub-sampled to 25%, 10%, 5%, and 1% of the original counts by strided sub-sampling across the ten minute dataset (rather than taking continuous, short-duration portions of list mode data). Reconstructions were performed with and without the application of motion correction (motion transforms estimated from external camera applied to correct individual lines of response).
Results: For a 10-minute PET bed in which the subject was instructed to hold still, a maximum of 1.5 degrees of rotation and 4 mm of displacement was observed. Large rotations and translations correlated with start and stop times of MR sequences, while small periodic motions related to respiration are observed in between. Full dose uncorrected PET images were slightly blurry compared to motion corrected (MC) images, but were still diagnostic. Qualitatively, uncorrected images showed poor gray/white contrast as dose was reduced, while contrast and features were preserved with the application of MC. Average root mean squared error (RMSE) of PET values in brain, relative to full dose MC, were 1.705 x103 Bq/cc for 25% dose, 3.014 x103 Bq/cc for 10%, 4.536 x103 Bq/cc for 5%, and 9.913 x103 Bq/cc for 1% with MC applied. In comparison, the RMSE of uncorrected images was 4.480 x103 Bq/cc for 100% dose, 4.785 x103 Bq/cc for 25% dose, 5.252 x103 Bq/cc for 10% dose, 5.918 x103 Bq/cc for 5% dose, and 10.129 x103 Bq/cc for 1% dose. Peak signal to noise ratios (PSNR) for MC images were 0.9540 for 25%, 0.8827 for 10%, 0.7981 for 5%, and 0.5062 for 1%. PSNRs of uncorrected images were 0.8175 for 25%, 0.7672 for 10%, 0.7090 for 5%, and 0.4924 for 1%.
Conclusions: To realize the full potential of low dose PET imaging, a system of motion management is necessary due to increased examination times. By integrating an optical tracking system, motion can be managed in a robust way. Future development of low dose imaging is likely to benefit from the application of deep learning algorithms. However, these approaches are unlikely to be successful without the integration of motion data. Therefore, retrospective motion correction enabled by using event-by-event list mode reconstruction, can enable the best quality low dose images, which is able to efficiently use all counts in reconstruction.
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.