(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: 3D images during treatment delivery can help assess the dosimetric impact of intrafractional motion and enable treatment plan adaptation to account for deviations in the delivered dose. This study sought to develop a method that enables the generation of time-resolved intrafractional 3D images from individual real-time kV radiographs acquired at limited angles during treatment.
Methods: 4DCT and intrafraction kV radiographs were collected for 3 patients. A mid-position (MidP)-CT was first generated from the planning 4DCT. Using deformable image registration (DIR), the MidP-CT was registered with each of the 10-phase images of 4DCT. The motion state M of any anatomical structure can be modeled with M = S + a·Df, where S is the reference state (MidP), a is the amplitude, and Df represents the output obtained by interpolating the input DVFs in correspondence of phase f. 3D images corresponding to each motion state were used to generate a DRR database. Real-time kV radiographs acquired during treatment delivery were compared to the DRRs in the database to determine the a and f of the motion model which was in turn used to generate a 3D image.
Results: The computational model was developed for 3 lung cancer patients. For each patient, DIRs were verified by 300 landmarks. Results showed that the mean registration errors were 2.1±1.1, 2.4±0.9, and 2.9±0.8 mm and the percent intensity difference between radiographs and best-matched DRRs pixel intensities over all pixels was 6.5±3.6, 6.1±4.4, 7.6±4.1% for three patients.
Conclusion: real-time deformable model has been integrated with real-time 2D kV radiographs to generate intra-fractional 3D images. DRRs generated from these images showed a good agreement in pixel intensities with correspondent radiographs. The reconstructed 3D images can be used for intra-fractaional dose accumulation for adaptive radiotherapy in the presence of inter- and intra-fractional motion for thoracic and abdominal treatments.
Keywords
Anatomical Models, Registration, Reconstruction
Taxonomy
Contact Email
