Measurement of Prostate Inter-Fraction Movement by Tracking Implanted Fiducials in Daily CBCT Projections

Abstract

To develop a method for accurately determining implanted fiducial locations in 3D using template matching and sequential stereo in CBCT projection data. This technique will be used in both retrospective studies to improve motion models and to improve fiducial tracking during treatment for motion management. During planning each fiducial is contoured individually. Using non-clinical research software, a separate template is generated from the planning CT for each fiducial. In offline mode each fiducial is tracked in the CBCT projections by template matching followed by sequential stereo. In the offline mode the full scan data is available so the sequential stereo is applied in retrospective analysis mode (as opposed to real-time tracking), and by iteration outliers are removed. The match score threshold used for accepting 2D matches is initially set to a high value and reduced incrementally until a minimum of 100 trajectory points (out of > 900 projections) are produced with a range of motion below a threshold. The estimated 3D location is taken as the average of results from these trajectory points. Horn’s method for absolute orientation using unit quaternions is used to determine the 3D translation and rotation of the fiducials from the planning CT scan to the CBCT. Results from a prostate patient, with 3 implanted fiducials and 9 treatment CBCTs are shown here. The tracked fiducial locations were overlaid on the CBCT scan and evaluated by an observer. No failed detections were identified in this preliminary dataset. The recorded prostate motion was random over the course of treatment, with no clear temporal trends. The largest translations were in the Z (P-A) direction and rotations about the X (L-R) axis. Amplitude of motions were within expected range. The residual mismatch in fiducial positions after rigid 3D transformation was on the order of the voxel size, or less. CBCT projection images can be used directly for 3D fiducial localization. Further analysis is being carried out, but the technique may benefit from not being limited by the CBCT voxel size, not influenced by CBCT reconstruction artifacts and more robust to motion. The technique can be used in both retrospective applications, where projections are archived, and online for motion management.Tabled 1Abstract 2732; TableDays Post SimulationTranslation IEC (mm)Rotation IEC isocenter (deg)TxTyTzRxRyRz13-0.042.17-3.1412.02.72.316-0.67-0.20-0.435.13.80.9300.322.25-8.3124.01.81.7370.600.25-0.196.82.81.7442.710.32-0.097.42.52.6510.701.86-6.2720.04.43.851-0.57-4.064.21-15.53.4-2.061-0.622.98-6.2821.24.42.864-1.642.892.248.35.40.4Mean0.090.94-2.039.933.461.57Range-1.6, 2.7-4.1, 3.0-8.3, 4.2-15.5, 241.8, 5.4-2.0, 3.8 Open table in a new tab