Comparison of Accuracy and Stability of Target Position Prediction Between External- and Internal Surrogate-Based Correlation Model
Y Sakurai1*, M Nakamura1, H Iramina1, Y Iizuka1, T Mitsuyoshi1, Y Matsuo1, T Mizowaki1, (1) Kyoto University
Presentations
PO-GePV-M-189 (Sunday, 7/10/2022) [Eastern Time (GMT-4)]
ePoster Forums
Purpose: To compare accuracy and stability of target position prediction between external- and internal surrogate-based correlation model.
Methods: Nineteen patients with lung cancer who underwent two or three dual-source cone-beam computed tomography (CBCT) scans were included. A total of 702 projection images was acquired for CBCT scanning of 70 s. The sampling interval was 0.2 s. The motion of abdominal surface, which was used as an external surrogate, were acquired during scanning. In contrast, the motion of diaphragm was acquired as an internal surrogate after detecting the edge of diaphragm on each projection image. In addition, three-dimensional (3D) position of the implanted fiducial markers around the tumor were determined as ground truth. The regression-based correlation model between the target and surrogate was constructed to predict target position with training period of 10 s of the first scan. Model parameters were not changed. The same procedure was applied to time-reversal data for data augmentation. The 3D prediction error, which was defined as median of 3D difference between predicted target position and ground truth, was calculated. Prediction accuracy was calculated as the 3D prediction error in the rest 60 s of the first scan (short-term evaluation) and 70 s of the last scan (long-term evaluation). Prediction stability was defined as the interquartile range of absolute difference between 3D prediction error of short- and long-term evaluation.
Results: The median time between first and last scan was 6.5 min (range: 2.5-9.0 min). The medians of prediction accuracy in short-term evaluation were 1.35 and 1.29 mm for external and internal surrogate, respectively. Those in long-term evaluation were 1.92 mm for both external and internal surrogate. Prediction stability were 0.22-2.50 mm and 0.15-1.34 mm for external and internal surrogate, respectively.
Conclusion: Internal surrogate is potentially more stable to predict target position than external surrogate.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email