Program Information
Three-Dimensional Target Motion Estimation From Dual-Source CBCT Projections and Extracorporeal Infrared Marker
H Iramina1,2*, M Nakamura2 , Y Iizuka2 , T Mitsuyoshi2 , Y Matsuo2 , T Mizowaki2 , I Kanno1 , (1) Kyoto University, Graduate School of Engineering, Kyoto, Kyoto, (2) Kyoto University, Graduate School of Medicine, Kyoto, Kyoto
Presentations
SU-H4-GePD-J(A)-5 (Sunday, July 30, 2017) 4:30 PM - 5:00 PM Room: Joint Imaging-Therapy ePoster Lounge - A
Purpose: To estimate unidentifiable target positions from CBCT projections.
Methods: We have developed dual-source 4D-CBCT imaging system with extracorporeal 1D infrared marker (IRM). Our proposed method to estimate unidentifiable 3D target positions was as follows; (1) Training session: constructing an estimation equation of identified 2D target positions on projections with IRM positions and velocities which measured by the independent system (2DEstEq). These positions were also converted into 3D positions to construct another estimation equation (3DEstEq). (2) Estimating session: estimating 2D or 3D target positions from IRM positions and each constructed EstEq. To validate the method, we used the data of 351 2D positions from each source and 351 converted 3D positions of the implanted marker for sixteen lung cancer patients. The detected 3D positions were used as actual positions in this study. We estimated 3D positions in latter half (176) of data from 2DEstEq and 3DEstEq made by former half with IRM (2D-3D and 3D-3D estimation, intra-scan estimation). Since 15 out of 16 patients underwent two or three CBCT scans, the validation under the existence of median elapsed time of 10 min between sessions was also employed (inter-scan estimation). Root-mean-square-errors (RMSEs) were evaluated between actual and estimated LR, SI, and AP positions.
Results: Median peak-to-peak target displacement in LR, SI, and AP directions were 2.9, 18.3, and 5.4 mm, respectively. In intra-scan estimation, median RMSEs of 3D-3D estimation were 0.4, 1.8, and 0.8 mm, in LR, SI, and AP directions, which were significantly lower than those of 2D-3D estimation except for SI direction. In inter-scan estimation, median RMSE was increased by 1.3-fold in SI direction due to baseline drift between sessions.
Conclusion: Unidentifiable 3D target positions were well estimated from IRM position by our proposed method; however, baseline drift should be managed to improve estimation accuracy.
Contact Email: