Abstract
BACKGROUND: Accurate tumor localization is crucial in radiation therapy to ensure precise dose delivery and minimize damage to healthy tissues. This study introduces a novel thoracoabdominal phantom designed for predicting tumor positions in radiotherapy. The phantom incorporates the use of mask region-based convolutional neural networks (Mask R-CNN) ultrasound image tracking algorithm (M-UITA) in conjunction with 4-dimensional computed tomography (4DCT) to establish and refine a tumor motion conversion model. METHODS: Respiratory Motion Simulation System (RMSS) along with 4DCT was used to track the motion trajectories of the tumor phantom in both the superior-inferior (SI) and medial-lateral (ML) directions, with amplitudes ranging from 30-40 mm. Simultaneously, M-UITA was used to track the motion trajectory of the diaphragm phantom in the SI direction to establish a conversion model to derive the motion of the tumor from the motion of the diaphragm. Subsequently, cone beam computed tomography (CBCT) was used for the verification of the tumor phantom conversion position error. RESULTS: The results indicated that the absolute error between the estimated and actual motion trajectories of the tumor phantom ranged from 0.35 to 1.35 mm in the SI direction and from 0.73 to 2.26 mm in the ML direction. CONCLUSIONS: This study has redesigned the thoracoabdominal phantom and refined the conversion model. In comparison to previous research, errors in both the SI and ML directions have been reduced. In the future, it can be integrated with a respiratory motion compensation system to minimize radiation dose damage to normal tissues.