Abstract
578
Objectives: Respiratory gating has been used in PET imaging to reduce the amount of image blurring caused by patient motion. However, segregating data into separate gated bins reduces the count statistics in the images and thus introduces risk of degraded diagnostic quality. Furthermore, there are no established protocols for physicians to interpret 4D data sets and thus it is difficult to integrate 4D imaging in clinical operations. Sorting data into a single optimal bin has been proposed as a solution - this option has the advantages of being simple to understand (simply using a time-subset of data), and results in a single 3D image for physicians to read. To date, optimal binning protocols have been built upon externally driven motion characterization, and population derived optimal binning strategies. In this work, we are proposing a new strategy of characterizing patient motion directly from the patient scans, and using that signal to determine a patient specific optimal bin.
Methods: A population of 219 FDG PET scans were studied. 4D respiratory gated PET datasets were generated using data driven gating and phase based binning (10 bins per cycle). For each scan, a data driven amplitude motion characterization was generated using principle component analysis of the 4D data set. Specifically, a phase-amplitude relationship was described for each scan by plotting the 1st principle component (PC) weight array against phase. In this way, an amplitude of motion characterization was derived from a phase gated data set. Next, a patient specific “optimal bin” was derived by comparing the 1st and non-1st PC weight arrays - the non-1st PC arrays were used to delineation the significant/non-significant changes in the 1st PC data. The optimal bins were determined by identifying the phases that had the maximal number of statistics included within the significant changes window.
Results: 219 gated PET scans were processed and each binned into a single, optimal, 3D PET image. For 69% (n=152) the optimal bin was determined to include 100% of the image statistics, i.e. the patients did not noticeably benefit from the gating effort. In the remaining images, the optimal statistics binning windows were an average of 65% of the statistics, and ranged between 20%-90%. 98% of images assessed as not having motion, through agreement with 3 blind reviews, were presented as optimal using 100% statistics - qualitatively validating the algorithm. Visual inspection of the optimally binned images consistently showed better resolution in the optimally binned images when compared to the non-gated images, and better noise characteristics than the original phase gated images (which by definition were noisier).
Conclusion: Conformal methods of processing are especially appropriate for the area of 4D imaging, as patient motion and signal statistics are both patient specific and scan specific. They can address a significant issue that many patients do not benefit, or benefit minimally, from the effort of gating, and we need motion correction strategies that do not degrade their care. Fully automated data driven strategies come with many practical advantages, as they are easy to implement and operator independent. To date the majority of literature for optimal binning of data is based on population averages and is derived from external motion characterization signal. The method we presented is data driven and is conformal to the available signal existing in the image being read. In the end it outputs an easy-to-use 3D motion corrected image, in the same format as the nongated images that physicians are used to reading. In the future we plan to continue to develop this methodology, quantify its impact, and asses the applicability of integrating it into clinical operations. Research Support: We would like to thank Siemens AG for help in supporting the acquisition of the gated PET scans.