Inter-pass motion correction for whole-body dynamic parametric PET imaging

Objectives: Whole-body dynamic PET imaging can be achieved through multi-pass continuous bed motion (CBM) acquisition protocol on Siemens scanners. However, body motion during the scanning could cause inter-pass mismatch, subsequently affecting parametric image generation. In this work, we implemented a non-rigid registration method to correct inter-pass motion for dynamic PET.

Methods: 27 subjects (5 healthy and 22 cancer patients) were included in this study. Each underwent a 90-min whole-body dynamic FDG CBM PET scan on a Biograph mCT (Siemens Healthineers, Knoxville, TN). A single-bed scan over the heart was acquired first for 6 min and then 19 CBM passes (frames) were acquired (4 × 2 min and 15 × 5 min). The inter-pass motion correction was achieved through non-linear image registration with B-spline free-form deformations via BioImage Suite. Each 5-min frame was registered to frame 12 as the reference image. A multi-resolution optimization was performed first at a coarse level and then refined to higher resolution, using normalized mutual information as the similarity metric. All dynamic frames were converted to standardized uptake values (SUV) unit, and an intensity cutoff at an SUV threshold equal to 2.5 was applied when computing the displacement fields to minimize the impact from the high-intensity bladder voxels. The parametric images were then generated by fitting the Patlak model (t* = 20 min) using the individual subject’s arterial input function. The overlaid Patlak plot slope K_i and intercept V_b were visualized to identify any mismatch. 57 regions of interest (ROI) (8 benign and 49 malignant) were drawn on hypermetabolic hotspots in 22 cancer subjects. The normalized fitting errors were compared using the paired two-tailed t-test for the whole body, head, and ROIs. In K_i images, we conducted ROI analysis and estimated the classification capability between benign and malignant ROIs using a receiver operating characteristic (ROC) curve and the area under the curve (AUC).

Results: The alignment between K_i/V_b images improved after inter-pass motion correction. Voxel-wise normalized fitting error images showed global error reduction with motion correction. Quantitatively, the normalized fitting errors significantly decreased in the whole body (p = 0.0496), head (p = 0.0090), and ROIs (p = 0.0022). In both benign and malignant ROI groups, the mean values for each ROI decreased after correction while the maximum values and standard deviation increased. The AUC of mean K_i after motion correction increased from 0.801 to 0.806.

Conclusions: The inter-pass motion correction using non-rigid registration has a great potential to enhance the alignment between K_i and V_b images, and improve the parameter estimation accuracy. Acknowledgments: This work was supported by NIH grant R01 CA224140.

• Download figure
• Open in new tab
• Download powerpoint