Real-time lung tumor tracking on MV bev images using a markerless model

Purpose: Real time lung tumor motion tracking allows for target dose escalation while simultaneously reducing normal tissue toxicity. We have developed a method that analyzes MV projections along with simulated digitally reconstructed radiographs (DRRs) as references to track tumor motion trajectories without implanting any fiducial markers. In this work, we successfully extended our analysis to phantom studies with small tumors with five motion patterns over 9 gantry angles. Additionally, we examined 5 SBRT lung cancer patients for the feasibility of critical dailyverification of tumor motion ranges. Methods: This algorithm divided the global composite DRRs and MV projections correspondingly into matrix of sub-images called tiles. Registrations were performed on tile pairs independently to take advantage of local soft tissue contrast and reduce global image discrepancies due to different imaging modalities. This algorithm is evaluated by phantom studies. A simulated tumor was controlled to move in a complex humanoid torso within a range of 10 mm to 20 mm that included the heart, ribcage/chest-wall bone, skin and sub-dermis. Treatment beam images were acquired continuously in beam\u27s-eye-view during the delivery of nine-field 3D conformal treatment plans to the phantom. Results: Approximately 7500 projections were acquired and analyzed for phantom studies with a tumor size of 6.9 cm3 at multiple tumor sites. Tumor positions were identified on every projection with a maximum/average error of 1.8 mm/1.0 mm. This algorithm has also been successfully applied to over 5000 frames of MV projections acquired during radiation therapy of 5 lung cancer patients. Conclusion: The accurate characterization of intra-fractional lung tumor motion, without the need for implanting internal fiducial markers provides critical daily-verification of whether the tumor motion range is within prior planning estimates and covered by the treatment planning volume. It could be applied for online treatment monitoring for treatment plan delivery with reduced planning margins