Abstract
P285
Introduction: Patient respiratory motion can degrade the quantitative accuracy on PET/CT images. The purpose of this study was to evaluate the image quality of our new respiration-gated image reconstruction algorithm which automatically generate a single-gate image with the least effect from respiratory motion at PET/CT imaging.
Methods: We performed a phantom study and a retrospective analysis of 14 patients with liver metastases (lesions n = 48) who underwent FDG-PET/CT imaging. We acquired listmode data using a commercially available PET/CT scanner (Cartesion Prime, Canon Medical Systems, Otawara, Japan). acquired for Whole-Body study was divided into very short time frames and reconstructed. Then, images were compressed into feature vectors for the computation efficiency. By applying Principal Component Analysis on the series of feature vectors, we obtained the respiratory motion waveform. We used the waveform and feature vectors to conduct two types of respiratory gating images, phase-gating and auto-gating. For the phase-gating, we processed the obtained waveform as the same fashion as the conventional respiration monitor device-based phase-gating. For the auto-gated reconstruction, we extracted the 35% of total time frames by comparing the similarity of feature vectors. We adopted the Euclidian distance of feature vectors to measure the similarity. The collection of similar data is expected to yield the image with the least effect of respiratory motion. In our phantom study, we filled a NEMA IEC-body phantom with FDG solution of activity contrast 1:4 (background: spheres). The phantom was placed on the driving unit that moves by following the input waveform signal. We used the waveform taken from a real human respiration of about four seconds per cycle, but the last 40% of the waveform was modified to mimic the change of breathing depth. Data were obtained at five reconstructions: stationary (for reference), no gating with respiratory motion (non-gated), device-monitoring with respiratory motion (device), Data-Driven-Phase-gating (phase), and Data-Driven Auto-gating (auto). For the 10 mm sphere, the SUVmax and the mean squared error were calculated for each PET Reconstruction. In the clinical study, PET images were obtained using three methods: non-gated-, phase-, and auto reconstruction. We measured SUVmax, SUVpeak, and diameter of the lesions on each reconstructed image.
Results: In the phantom study, SUVmax (mean squared error) values were 4.97 (0.0) for reference-, 1.44 (20.8) for non-gated-, 2.59 (13.8) for device-, 2.46 (14.6) for phase-, and 4.68 (5.10) for auto data. Auto- were closest to the reference data. In the clinical study, the median SUVmax (interquartile range) was 6.22 (3.76 - 10.26) for non-gated-, 6.44 (3.79 - 11.29) for phase-, and 7.13 (4.10 - 11.77) for auto data. The median SUVpeak value was 3.95 (3.04 - 6.11) for non-gated-, 4.14 (3.15 - 6.94) for phase-, and 4.32 (3.25 - 6.71) for auto data. With respect to SUVmax and SUVpeak there was a statistically significant difference (p < 0.05) between phase- and non-gated-, and between auto- and non-gated data although there was no significant difference between phase-, and auto images (P=0.28, 0.20, respectively). The median tumor diameter (interquartile range) was 14.8 mm (12.16 - 19.03) on non-gated-, 12.68 mm (10.57 - 16.91) on phase-, and 10.57 (quartile 8.46 - 16.91) on auto images. It was significantly smaller on auto- than on non-gated- and phase images (p < 0.05).
Conclusions: Our new auto-gated reconstruction algorithm for PET/CT imaging improved both the image quality and the quantification of SUV values
In this issue
Jump to section
Related Articles
Cited By...
- No citing articles found.