ESTRO 2023 - Abstract Book

S238

Saturday 13 May

ESTRO 2023

1 University Medical Center Mannheim, Department of Radiation Oncology, Mannheim, Germany; 2 Universidad Rey Juan Carlos, Medical Image Analysis and Biometry Lab, Móstoles, Spain; 3 GMV, Healthcare, Madrid, Spain; 4 University Medical Center Mannheim, Mannheim Cancer Center, Mannheim, Germany Purpose or Objective Optical tracking in combination with preoperative imaging provides the opportunity to navigate an x-ray source in intraoperative radiotherapy (IORT). The knowledge of the x-ray source isocenter represents the basis for treatment planning. Tracking the isocenter with an optical tracking system and a rigid body with optical markers requires a previous calibration to define the offset from the rigid body to the isocenter of the source. The aim of this project is to create a navigation workflow, which yields an optimized x-ray source positioning and allows for treatment planning for kypho-IORT. Materials and Methods Two rigid bodies with four optical markers each were used. One was connected to the floor stand of an Intrabeam system (Carl Zeiss Meditec AG) to track the x-ray source and the other one was fixed to a needle applicator to be used as calibrator with a known offset between its markers and the source isocenter. Measurements were taken with a Polaris Vega system (Northern Digital Inc.). The calibration was performed, recording movement in 750 frames. In order to simulate a treatment navigation, three 3D printed vertebras were used and seven metallic fiducials were attached for registration purposes. Next, another needle applicator was connected to the floor stand and placed close to the vertebras and a cone beam CT (CBCT) was performed. The fiducials’ positions previously recorded in the optical tracking system and determined in the CBCT scan were used to register both systems. Finally, the isocenter coordinates in the CBCT scan and in the optical system were determined and evaluated regarding their geometrical deviation. Results The calibration estimating the offset between rigid body of the floor stand and isocenter showed a mean geometrical error of 0.24 ± 0.13 mm and 0.65 mm as maximum error. Five trials of the navigation were evaluated and the results are shown in Table 1. For the registration from the optical system to CBCT scan, the fiducial registration error (FRE) was between 0.87 mm and 1.04 mm. Comparing the isocenter coordinate from the CBCT and the optical tracking system, the geometrical deviation was between 0.78 mm and 1.58 mm. Table 1: FRE and error of the isocenter coordinate for each trial Trial FRE (mm) Isocenter coordinate error (mm) #1 0.98 1.58 #2 1.04 0.82 #3 0.92 1.07 #4 0.87 0.78 #5 0.89 1.22 Conclusion In this project, we showed the feasibility to use an optical tracking system to navigate an x-ray source for kypho-IORT. The influences of different equipment used during a surgery in an operating room and of the patient ifself on the precison of the navigation need to be evaluated in further studies. MO-0304 Privacy-preserving federated learning for radiotherapy applications H. Hayati 1 , S. Heijmans 2 , L. Persoon 2 , C. Murguia 1 , N. van de Wouw 1 1 Eindhoven University of Technology, Department of Mechanical Engineering, Eindhoven, The Netherlands; 2 Demcon Advanced Mechatronics B.V., Department of Software Engineering, Best, The Netherlands Purpose or Objective The use of AI models for personalized radiotherapy has been a technology trend for the last decade. This technology relies on training models from labeled patients' data distributed across different institutions. However, directly sharing patients' data across institutions during the training process raises privacy concerns, and the regulatory frameworks (e.g., GDPR) prevent these data exchanges. Federated Learning (FL) has emerged as a privacy solution that allows multi-institutional distributed model training by exchanging models' parameters instead of sensitive patients' data. In FL, local models are trained on local datasets and transferred to a server to aggregate a global model. Although FL can provide privacy to some extent by keeping patients' data locally, it has been shown that information about patients' data can still be inferred from the local models during the training process. Various privacy schemes have recently been developed to address this privacy leakage, however they all provide privacy at the expense of model performance or system efficiency. In this work, we propose a Privacy-Preserving FL (PPFL) scheme built on the synergy of matrix encryption and system immersion and invariance tools from control theory. We show that this scheme provides strict privacy guarantees for patients' data without compromising the accuracy and performance of the FL model. We demonstrate the performance of our tools on a FL model for chest radiograph interpretation (CheXpert dataset). The CheXpert goal is to predict the probability of 14 different observations from multi-view chest radiographs. Materials and Methods Mini-Oral: Audits and multi-centre studies