Data-Driven Respiratory Gating Whole Body Pet Using Continuous Bed Motion

Data-driven respiratory gating methods eliminate the requirement for hardware based devices used to respiratory gate PET studies, but few approaches can reliably extend to acquisitions greater than a single axial field of view. A fundamental limitation of many data-driven methods lies in the inherently arbitrary relationship between signal polarity and the physical direction of motion at different axial locations. We have produced a data-driven gating method which exploits continuous bed motion to overcome this issue. Listmode time-of-flight PET data was converted to a time series of spatially filtered histo-projection volumes, and an initial estimate of the respiratory signal was obtained by calculating the time-varying anterior-posterior (AP) displacement. The full acquisition range was then divided into a series of overlapping short axial regions and processed with a data-driven gating method based on spectral analysis, initialized with spectral information from the AP signal. An optimization process was used to combine the axial regions and produce a consistent relationship between the physical direction of motion and the respiratory signal polarity throughout the acquisition range. To produce gated images with axially uniform noise, an adaptive gating methodology was implemented to correct for temporal variations in the respiratory signal characteristics. We analyzed 86 patient acquisitions, and both methods produced similar results between the bladder and the aortic arch, with an average correlation between data-driven and hardware signals of 0.81 (+/- 0.1). Low correlation was frequently found in regions where little or no motion was present. In some cases where low or negative correlation was found, a larger extent of upper lung respiratory motion was identified in images gated with the data-driven signal, suggesting that hardware signals can potentially be less accurate in regions axially distal to the device itself.