Abstract
460
Objectives: Respiratory motion is a resolution limiting factor in thoracic PET imaging. For over a decade data driven motion correction (DDMC) has been studied as a practical solution, but has largely been limited in advancement because of the structural hurdle of sharing solutions across the imaging and research communities - DDMC has required access to proprietary data formats not easily shared. We are beginning to see changes in data access infrastructures, vendor support for data driven innovation, and increased interest in DDMC. Correspondingly our group has built a sharable DDMC solution to help support a new era of data driven PET research.
Methods: KesnerDDG is a fully automated command line executable that can be used to process a PET listmode file, or population of files, for motion correction. The specific software input is a listmode filename, and the output consists of (a) a new version of the listmode file with respiratory triggers in it (analogous to hardware) that can be used to support gated reconstruction, (b) a patient/scan specific respiratory motion characterization curve, and (c) a data driven (/scan specific) optimally binned subset of events in a triggerless listmode file. The software can serially process any number of listmode files specified by user, making it useful for population research. Users are given full control of 18 processing parameters to support further research on parameter optimization - including Kesner’s method or PCA-based method engines, both benchmarked in literature (1,2). The software was tested using data from 4 different scanners from two vendors, and on a population of 220 previously benchmarked FDG PET scans.
Results: The fully automated function of the software makes it an easy-to-use DDMC tool, that can be integrated with fully automated workflows. In addition, this code (optionally) integrates data driven motion characterization with data driven-optimal binning, giving users the ready ability to create a 100% fully automated motion correction workflow which results in an easy-to-read 3D PET image. This new unique workflow safeguards correction such that only patients who benefit from motion correction processing have their original (non-gated) image data modified, thus sparing image degradation from motion correction to the large subset of patients who don’t receive benefit, and does not require assumptions used in elastic transformations. The code has been tested on 230+ listmode files. It is anticipated that more vendor/scanner functionality will be added in future, in response to user interest. The mean/median/SD Pearson correlation of the respiratory characterization curve in the population of benchmarked data (1) and our new software-generated data were 0.82/0.98/0.30 and 0.63/0.73/0.32 for Kesner and PCA methods, respectively. The similar-but-different results are expected and indicate an initial general validation of the software, and an example of how differently performing methods could be compared. Further validation and analysis will be performed in the future.
Conclusions: Community interest in DDMC is increasing. KesnerDDG is a non-profit, free DDMC tool for generating motion characterization and data manipulation to address motion in PET. It can be used as an accessible tool for DDMC research, to benchmark other research/vendor products, to compare DDMC strategies across centers/scanners, and to encourage further competitive development. Ideally, this project it will also serve as a future platform for community research and collaboration. More information, including detailed information, instruction manual, instructional videos, and information on how to obtain a copy via research agreement, can be found online: www.kesnersmedicalphysics.com. ACKNOWLEDGMENTS: We would like thank GE and Siemens for their support in accessing scanner data. We would like to thank Dr. David Lynch (National Jewish Health) for sharing their research data set.