Abstract
3279
Introduction: Head motion during PET neurology scans leads to image blurring and reduction of quantitation. We previously presented a data-driven motion correction approach for the brain and a corresponding clinical evaluation. However, the quantitative “truth” in clinical datasets cannot be determined. In this work, we perform a phantom study to evaluate the quantitative accuracy of the data-driven motion correction approach, even in the presence of large motion.
Methods: The Hoffman 3D Brain Phantom (Data Spectrum, Durham, North Carolina, USA) was filled with 79 MBq (at scan start time) of 18F solution. A sphere of inner diameter 16.5 mm (2.3 mL volume) was filled at a 4:1 ratio to the phantom background liquid and inserted within the brain portion of the phantom. The phantom was positioned within the 8-channel brain coil on a GE SIGNA PET/MR scanner.
Acquisition was performed for a total of 80 min, with varying magnitudes and types of motion. For the first ten minutes, the phantom was not moved in order to obtain a “gold standard” image without motion. Both step-wise and continuous motions were performed, with rotations up to approximately ±30° and translations up to approximately 25 mm.
Data-driven motion estimation was performed over the full acquisition duration using 3D reconstructions from short data frames (each frame < 1 s), registered to a reference frame at the initial phantom position. After motion estimation was complete, the data were split into 8 frames (each of 10-min duration) to investigate the effect of varying magnitudes of motion. For each frame, reconstruction was performed without motion correction (No MoCo) and with motion correction (MoCo). Additionally, No MoCo and MoCo reconstructions were performed over the full 80-min duration for comparison. All reconstructions were performed in list mode. For the MoCo reconstructions, the position of each projected coincidence event was adjusted based on the data-driven motion estimation. Motion-aware data corrections were applied for the MoCo reconstructions.
A volume of interest (VOI) was placed on the 16.5 mm lesion for each of the 8 frames, for both the No MoCo and MoCo images. The VOI mean was compared to the first frame of the No MoCo reconstruction (when the phantom was not moving). Additionally, difference images and image profiles were created and inspected for potentially problematic quantitative gradients or offsets in the MoCo reconstructions.
Results: Comparison between the 8 MoCo reconstructed frames and the No MoCo reconstruction from the “gold standard” frame (the first 10 min without motion) demonstrated consistent quantitation. No quantitation issues were observed that would be considered clinically problematic. Qualitative comparison of the MoCo images demonstrated success of the motion correction algorithm.
The VOI mean measurements of the lesion were consistent within < 2% for all MoCo frames, as compared to the No MoCo “gold standard” frame.
Conclusions: Quantitative accuracy of the data-driven motion correction algorithm was demonstrated through this phantom study, even for large degrees of continuous motion.
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.