Program Information
A Bayesian Approach for 3D Markerless Lung Tumor Tracking Using KV Imaging
C Shieh1*, V Caillet1,2 , M Dunbar1 , P Keall1 , J Booth2 , N Hardcastle2 , C Haddad2 , T Eade2 , I Feain1 , (1) University of Sydney, Camperdown, NSW, (2) Royal North Shore Hospital, Saint Leonards, NSW
Presentations
SU-E-601-1 (Sunday, July 30, 2017) 1:00 PM - 1:55 PM Room: 601
Purpose: The ability to monitor tumor motion without implanted markers can enable broad access to more accurate and precise lung radiotherapy. kV-based markerless tumor tracking on existing systems faces two challenges: the inferior tumor contrast on kV projections, and the inference of 3D tumor position from 2D imaging. The aim is to develop and clinically validate a single-projection-based 3D markerless lung tumor tracking method for the first time.
Methods: The proposed method consists of a model-based contrast enhancement and Bayesian 2D-3D inference. For each intrafraction kV projection, the tumor contrast is first enhanced by removing the attenuation contribution of background anatomic structures using a 4D patient model built from the pre-treatment CBCT. The 2D tumor position is then measured by template matching. A Bayesian 2D-3D inference framework is used to first predict the most likely 3D tumor position based on prior knowledge of the patient’s respiratory motion. The measured 2D and predicted 3D positions are combined to yield the optimal 3D estimate based on template matching uncertainty and past positional distribution. This method was retrospectively validated on 13 treatment fractions from 6 locally-advanced or SABR patients. Tracking errors were estimated using the motions of markers or beacons implanted near the tumors. To avoid biasing the tracking results, markers and beacons were removed from the kV projections.
Results: The mean and 95th-percentile of 3D tracking errors were 1.6-2.9 mm and 2.6-5.8 mm. Tracking errors in the left-right, superior-inferior, and anterior-posterior directions were -0.05±1.05 mm, -0.50±1.74 mm, and 0.13±1.49 mm (mean±SD). Compared to no tracking, markerless tumor tracking enabled an average margin reduction of 3.9 mm and up to 9.5 mm.
Conclusion: The feasibility and benefits of 3D markerless tumor tracking on existing systems were demonstrated for the first time. Its clinical implementation can enable widely accessible real-time adaptive lung cancer radiotherapy.
Contact Email: