ESTRO 2021 Abstract Book
Conclusion During PBHs, considerable variation in motion occurs across the diaphragm. D2 motion highly correlates with that of the right kidney. As the established method accounts for varying degrees in anisotropy and heterogeneity of motion, it is a promising tool for organ motion quantification during PBHs. PH-0265 Diaphragm motion prediction based on optical surface with machine learning for liver tumor SBRT Z. Dai 1 , Y. Zhang 2 , Q. He 3 , S. Zhao 3 , Y. Zhu 3 , H. Jin 3 , J. Chen 4 , X. Wang 3 1 The Second Affiliated Hospital of Guangzhou University of Chinese Medicine, Department of Radiation Therapy, Guangzhou, China; 2 National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital ， Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, Department of Radiation Therapy, Shenzhen, China; 3 The Second Affiliated Hospital of Guangzhou University of Chinese Medicine, Department of Radiation Therapy, Guangzhou, China; 4 Guangzhou University of Chinese Medicine, School of Medical Information Engineering, Guangzhou, China Purpose or Objective To establish a correlation model between the body surface and the diaphragm, and realize the automatic prediction of the superior-inferior (SI) motion trajectory of the diaphragm based on the 3D body surface information from the optical surface monitoring system ( OSMS ). Materials and Methods Six patients with liver tumor were enrolled in this study. We captured the body surface motion parameters by AlignRT (Vision RT Ltd, London, UK) and kV fluoroscopic image by Varian Edge treatment machine (Varian Medical Systems, Palo Alto) simultaneously. The correlation models between the body surface and the diaphragm were developed with two machine learning methods: linear regression (LR) and random forest (RF) based on synchronous monitoring data of internal/external motion for 60 seconds before radiotherapy for each patient. The locations of diaphragm centroid and apex were extracted according to the contour of the diaphragm manually delineated by the physician which was used for ground truth. The predictive performance of the model was evaluated retrospectively on clinical datasets. Model 1 was trained with the data for the first 44 seconds and the data for the last 16 seconds was used for testing during the first fraction in order to evaluate the intra-fractional prediction accuracy . During second fraction, Model 2 was constructed with the data for the first 44 seconds. The last 16 seconds motion trajectory of diaphragm was predicted with model 1 and model 2 respectively to evaluate the inter-fractional prediction accuracy. The prediction errors of the two models were compared to analyze whether the model need to be re-established for inter-fractional motion management.
Made with FlippingBook Learn more on our blog