Respiratory Motion Correction in Long Axial FOV Scanners

Introduction: Commercially available PET/CT scanners with long axial field of views (LAFOV) offer very good time-of-flight (TOF) resolution about 215 ps. The raw data sizes increase significantly with such scanners, nevertheless, currently achievable TOF resolution allows substantial data compression, the histo-projections data format in Vision PET/CT scanners. Furthermore, patient motion during the scan is unavoidable; however, non-rigid motion estimation and correction can be performed directly in the quasi-image space of histo-projections by exploiting the spatial information of the TOF signal properties. Specifically, an approximation of rigid motion on a scale of the TOF resolution is used. In addition, data driven motion detection is particularly attractive in LAFOV scanners since the entire patient is visible at once. Once motion correction is performed in histo-projections, a regular reconstruction can be performed on a single data set, in contrast to motion framed approaches. The current work focusses on respiratory Motion Correction (MC) using a Biograph Vision Quadra scanner that offers a 106cm axial FOV.

Methods: In this proof of concept study, initial patient data acquired on Biograph Vision Quadra was analyzed. A portion of the histo-image (that is combination of all view dependent histo-projections), significant in axial extend and covering the lung and abdomen regions, was used to obtain a gating signal in form of value of axial component of histo-image center of mass. A clear respiratory signal was derived in patients receiving FDG and PSMA. A patient with colon cancer and significant liver motion was chosen for further analysis (258 MBq of FDG with a 120 min uptake, single bed (106cm) acquisition of 4 mins duration) The reconstruction employed a maximum ring difference limited to 18° acceptance angle, the current clinical reconstruction protocol. However, the histo-image formation used the maximum ring difference, corresponding to all lines-of-response or 52°. The histo-image data were gated, resulting in four histo-image frames. Motion vector field was estimated from these gated histo-images by the Lucas-Kanade optical flow method.

Results: Improvement by removing motion blur and maintaining similar noise level was apparent in comparison between image space MC reconstruction and static (no MC) reconstruction. MC reconstruction background cardiac region was 14% noisier; the SUV of a liver lesion was about 80% higher in value compared with the SUV value using a static reconstruction. Comparing with single gate of a “motion frozen” reconstruction, MC reconstruction displayed less noise while maintaining contrast recovery: gated data reconstruction cardiac region was noisier about 50% and liver lesion was higher by 15%.

Conclusions: The standard optical flow motion estimation algorithm is capable of distinguishing respiratory motion in relevant body regions, such motion in the liver and identifying the absence of motion in the rest of body, such as head. The presented methodology performs non-rigid motion correction in sufficiently high TOF resolution quasi-image space. Acknowledgment: The authors would like to thank Bern University Hospital’s Inselspital Department of Nuclear Medicine, the first site of Siemens Vision Quadra, for providing patient data

• Download figure
• Open in new tab
• Download powerpoint