Abstract
166
Objectives: Respiratory motion during PET acquisition often causes image blur and inaccurate quantification of radioactivity concentration. Gated PET is traditionally used to correct these motion artifacts. However, gated PET often produces noisy images due to the fact that only a portion of the data is used in image reconstruction. The objective of this study is to utilize all acquired data of a gated PET scan toward reconstructing a motion-free image, such that motion artifacts are corrected without sacrificing image quality. This is achieved using our motion-incorporated reconstruction (MIR) with the help of a 4D CT scan.
Methods: Motion information is obtained from corresponding 4D CT scans by deformable image registration. Once the motion field between two time bins is determined, motion is modeled into an interpolation matrix, such that image vector of one time bin can be mapped to that of another by multiplication of the interpolation matrix. By treating the interpolation matrix as part of a “4D system matrix,” and by introducing a “4D sinogram,” the system model has the same form as that of a static scan, while having data from all time bins considered. As a result, a motion-free image can be reconstructed out of all collected data, and the reconstruction can be done using any existing iterative algorithms suitable for a static scan, among which the ML-EM algorithm was selected in our study. A phantom study has been performed on the GE Discovery ST PET/CT scanner to investigate the feasibility of this approach. The phantom contained a fixed and a moving hot sphere both placed in a warm background of water. A 4D CT scan was performed followed by a gated PET scan. The PET scan was performed in list mode for 30 minutes using 10 time bins. Three types of reconstruction were performed on the first 3 minutes of data: a static reconstruction, a gated reconstruction of the first time bin, and an MIR. In addition, a gated reconstruction of the first time bin on the full 30 minutes of data (3 minutes of equivalent data) was performed as the gold standard. An ROI was placed on the moving sphere at the location corresponding to the first time bin. Quantification error and signal to noise ratio (SNR) are defined based on the mean and the standard deviation of pixel values within the ROI.
Results: The image produced by the MIR exhibits closer appearance to the gold standard image, in both edge definition and noise properties. Using the MIR, the quantification error was reduced by 73% compared to gated scan, 94% compared to static scan, and the SNR was improved by 3 folds over the gated image.
Conclusions: The MIR we have presented showed a good potential in correcting motion artifacts while keeping the image quality close to that of a static scan of the same time duration.
- Society of Nuclear Medicine, Inc.