Impact of low coincidence count density on Quantification when using Elastic Motion Compensation in PET/CT imaging

Objectives: Recently Elastic Motion Correction (EMC) algorithms which compensate for respiratory motion in PET imaging while preserving all the acquired data have become commercially available. One such technique is EMC via deblurring (EMDB), which is also commercially known as ONCOFREEZE. This algorithm uses mass preservation optical flow registration to determine a blurring kernel between a single motion corrected image (HDChest 35% of the data) and a single non-motion corrected target image which uses 100% of the acquired data. The blurring kernel is then used in the EMDB reconstruction to suppress respiratory motion while preserving all the acquired data. The objective of this work is to determine how decreasing count density impacts the EMDB algorithm with respect to radiotracer quantification in a phantom evaluation.

Methods: A phantom simulating the motion of the abdominothorax with lesions(5 spheres of inner diameters of 10-28 mm) was scanned four times with 0, 1, 2 and 3 cm of motion amplitude on a Siemens mCT PET/CT with a continuous bed motion speed of 0.5 mm/s. To simulate acquisitions with lower acquired counts, new PET list datasets were created by randomly removing fractions of the data to retain 75%, 50%, 25%, 12.5%, and 6% of the original list dataset. Reconstruction of all list datasets was performed with the EMDB algorithm using a 35% duty cycle for the HDChest reference image, 2 iterations, 21 subsets, TOF, PSF, 200 x 200 matrix, and 5mm post filtration. Measurements of SUVmax and SUVpeak were made on all spheres, motion amplitudes and list datasets. Reconstructions with zero motion amplitude were considered as the gold standard for each respective group of list datasets with the same percentage of data. The ratio of SUVmax (RSmax) and SUVpeak (RSpeak) for each sphere, motion amplitude and list dataset was then calculated with respect to the gold standard. For list datasets with the same percentage of data, the average RSmax and RSpeak for each motion amplitude acquisition was taken across all spheres.

Results: As expected, RSmax and RSpeak decreased as motion amplitude increased. Furthermore, this decrease was larger when using smaller list datasets. In particular, the RSpeak values for 0, 1, 2, and 3 cm acquisitions were 1.00, 1.00, 0.88, 0.84 for the 12.5% list dataset and 1, 0.93, 0.78, and 0.75 for the 6% list dataset. Larger list datasets (25, 50, 75 %) all had values above 95% for RSmax and RSpeak.

Conclusions: While the EMDB algorithm uses all of the acquired PET data, its effect on PET image quantification depends on motion amplitude and count density with degraded results as the motion amplitude increases and the count density decreases (increased image noise). Caution should be used when interpreting the results of EMDB images reconstructed from low count acquisitions and large motion amplitudes.