Way To Estimate Mean Position, Motion Magnitude, Motion Correlation, & Trajectory Of Tumor From Cone-Beam CT Projections For Image-Guided Radiotherapy

Modern linear accelerators for radiation therapy are often equipped with a cone-beam CT (CBCT) system, which is essentially an x-ray source and imager that can rotate around the patient. A volumetric CBCT image of the patient anatomy at treatment is obtained by acquiring typically around 600 x-ray images during a full 360-degree rotation in 1 minute. The image acquisition is followed by comprehensive calculations that reconstruct the 3D image from the set of 2D projection images.

In many cases, one or more x-ray visible fiducial markers are implanted in the radiotherapy tumor target in order to make it visible on x-ray projection images. This technique is a standard method for prostate cancer, where it enables compensation of internal prostate displacements by image-guided prostate localization just prior to treatment.

When such fiducial markers are present, the projection images of the CBCT scan provide the 2D trajectory of the target during the CBCT acquisition in a rotating coordinate system. The actual 3D target trajectory can, however, not be reconstructed from this projected 2D trajectory without ambiguity since an infinite number of 3D trajectories results in the same projected trajectory. The aim of the present study was to develop a method that gives a good estimate of the actual 3D tumor trajectory from the projection images and to investigate this method for prostate and for abdominal and thoracic tumors.

The basic assumption of the method is that the target position can be described by a 3D Gaussian distribution. This 3D Gaussian distribution is unknown, but it is first estimated from the series of projection images by a maximum likelihood approach and then, in a second step, used to estimate the 3D target position for each 2D projection. The result is the estimated 3D target trajectory.

The proposed method was first investigated in a simulation study based on two large datasets of patient-measured tumor trajectories for prostate and for thoracic or abdominal tumors, respectively. The known 3D trajectories were projected onto a rotating imager as they would have been in a real CBCT acquisition. The 3D trajectories were then estimated from the 2D projections by the proposed method and compared with the actual 3D trajectories for calculation of the estimation error. These errors were typically much less than 1 millimeter.

Next, the method was investigated experimentally by acquiring CBCT scans of a Styrofoam block with an embedded fiducial marker that was manipulated by a programmable motion stage to reproduce a prostate trajectory and a lung tumor trajectory. The projected marker position was determined in each x-ray image and used to reconstruct the two 3D trajectories by the proposed method. As in the simulation study, the reconstructed trajectories agreed with the actual trajectory well within one millimeter.

Finally, clinical feasibility of the method was demonstrated by reconstructing the 3D tumor trajectory for a clinical CBCT scan of a patient with a fiducial marker implanted in a pancreas tumor. In this case, the trajectory estimation error could not be quantified because the actual tumor motion was unknown. But the estimated trajectory was very similar to the tumor motion as determined from a 4D CT scan (i.e. a time-resolved CT scan) of the patient.

The proposed method basically works because tumor motion along different axes is often correlated and because tumor motion is often very limited in some directions. The method determines these motion characteristics and exploits them when estimating the 3D tumor motion from the observed 2D motion. For prostate specifically, the method works because left-right prostate motion in general is very small and because cranial prostate motion often correlates with anterior prostate motion.

In conclusion, a method for accurate estimation of the 3D tumor trajectory at radiotherapy treatments has been developed. This 3D trajectory information is very valuable for a wide range of strategies for tumor motion management in radiotherapy.