Realtime Volumetric Image for Target Localization During Proton FLASH Radiotherapy
C Chang1, Y Lei2, T Wang3, J Zhou4, L Lin5, J Roper6, K Higgins7, T Liu8, J Bradley9, X Yang10*, (1) Emory University, Atlanta, GA, (2) Emory University, Decatur, GA, (3) Emory University, Atlanta, GA, (4) Emory University, Atlanta, GA, (5) Emory Proton Therapy Center, Atlanta, GA, (6) Winship Cancer Institute of Emory University, Atlanta, GA, (7) Emory Univ, Atlanta, GA, (8) Emory Univ, Atlanta, GA, (9) Emory University School Of Medicine, ,,(10) Emory University, Atlanta, GA
Presentations
WE-B-202-4 (Wednesday, 7/13/2022) 8:30 AM - 9:30 AM [Eastern Time (GMT-4)]
Room 202
Purpose: Proton FLASH requires robust image-guidance techniques to direct proton beams to targets due to large energy deposition in target regions compared to conventional treatment techniques. This study proposes a deep network to generate volumetric images through two orthogonal 2D X-ray images to localize the target.
Methods: The proposed feature mapping network includes three components for feature extraction from two orthogonal x-ray projections, feature re-alignment, and 3D estimation. A 2D convolutional layers-based feature extractor is used to obtain deep features from two orthogonal x-ray projections. Feature re-alignment is used to adjust the projections’ angles in a Cartesian coordinate system. The volumetric images are generated by a 3D convolutional layers-based 3D estimation subnetwork. Ten patient cases with 4D CT were used to demonstrate the proposed method. Each 4D CT includes ten respiratory phases, and the 2D orthogonal images are generated by forward projecting 3D CT from each phase to planes. Leave-phase-out experiments were performed to ensure that each phase amount of the 10 phases will be used for testing once. The evaluation metrics included mean absolute error (MAE), peak signal-to-noise ratio (PSNR), and structural similarity index metric (SSIM) for body contours.
Results: The quantities of each metric are averaged among all ten respiratory phases of each patient. The averaged MAE and PSNR are 55.98HU and 27.08dB with standard deviations of 1.91HU and 0.21dB for all patients. The averaged SSIM is 0.97±0.001 for body contours, which is sufficient to determine whether GTV is on the proton beam path during the treatment.
Conclusion: The proposed feature mapping network demonstrated that volumetric CT images could be inferred from two orthogonal kV projections. The proposed network potentially can provide a solution for in-treatment real-time on-board volumetric imaging for tumor localization and tracking to ensure the effectiveness of FLASH treatment.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email