Fiducial Marker Motion Tracking Using Dynamic Programming
H Wan*, J Ge, P Parikh, Washington University in St. Louis, MOTU-C-213CD-3 Tuesday 10:30:00 AM - 12:30:00 PM Room: 213CD
Purpose: To develop a novel algorithm based on a global optimization strategy - dynamic programming (DP) - to track the motion of fiducial markers due to respiratory motion in cone beam CT projections.
Methods: A two-step iterative algorithm was used to track the marker locations. First, search windows were constructed to limit the search space. Second, DP was used to minimize a cost function involving the marker location, trajectory, and intensity. This algorithm was tested on 11 human data sets with varying number (1-8), type (Visicoils, embolization coils, stents, surgical clips), size (0.1-2 cm), and shape of markers. Markers often overlapped with other markers or background objects. The algorithm was also compared to the traditional template matching algorithms.
Results: The 11 data sets contained 7144 projection images. The algorithm correctly identified the markers in 99.15% of the images. Of the images where the algorithm failed, the average error was 0.72+-0.39 mm. The algorithm takes about 0.1 seconds per marker per data set to run on a 1.6 GHz computer. Compared to template matching, which succeeded in only one of 11 data sets, our algorithm is more robust, faster, and requires no a priori knowledge about the markers.
Conclusions: Our work provides superior tracking capabilities of fiducial markers in cone beam CT projections. It can be used in clinical settings to aid radiation therapists in the patient setup process and gating window selection, decreasing the patient setup time and radiation exposure.
Research sponsored in part by Varian Medical Systems