ESTRO 2024 - Abstract Book

S3711

Physics - Dose prediction, optimisation and applications of photon and electron planning

ESTRO 2024

other hand, zero constraints and a random weight range from 90 to 100 were set for normal tissues. Finally, split all optimized plans by beams, and use the PB and MC dose calculation algorithms to calculate the single-beam dose distribution map, separately. By the presented data generating method, we used 9 cases to generate a total of 8064 paired data, 7168 pairs of data for training, and 896 for testing, and the patients in the test data are independent of the training data. Finally, the generated large amount of data is used for training in a multi-level U-Net neural network. The training process takes the single-beam dose distribution map of PB as input and the single-beam dose distribution map of MC as the label. Evaluate the advantages and disadvantages through the Gamma pass rate of the dose distribution map output by the model and the dose distribution map calculated by MC dose, and the gamma parameter is set to 1mm/1%.

Results:

The gdMC achieved high precision and efficiency, the average time cost was reduced from 5min to 3s and the 1mm %1 gamma pass rate between MC and gfMC dose was 97.2%0.2% for training dataset, 95.5%0.3% for testing dataset. However, the result generated through dose prediction is 88% at the gamma pass rate of 1mm/1%.

Conclusion:

The study successfully built a generalized fast MC algorithm, which produces MC-comparable dose, but has a significant time cost saving. Another significant innovation was the method of generating training data, the only 9 cases CTs can form 8064 paired training data. This allows a general model to be trained based on small batches of clinical data in the future, the study towards an accurate and efficient radiotherapy dose calculation.

Keywords: Dose calculation, Deep learning, generalizable

2800

Digital Poster

Boosting 4D robust treatment planning for IMRT

Niklas Wahl 1,2 , Remo Cristoforetti 1,2,3

1 German Cancer Research Center (DKFZ), Medical Physics in Radiation Oncology (E040), Heidelberg, Germany. 2 Heidelberg Institute for Radiation Oncology (HIRO), National Center for Radiation Research in Oncology (NCRO), Heidelberg, Germany. 3 University of Heidelberg, Faculty Of Physics and Astronomy, Heidelberg, Germany

Purpose/Objective:

Use of Planning Target Volume (PTV) margins to handle geometrical uncertainties is a well established procedure in radiation therapy. When considering 4DCT treatment plan optimization, common practice is to define an Internal Target Volume (ITV) as the combination of each CT phases' Clinical Target Volume (CTV). The PTV is then obtained by expanding the ITV volume by a given margin. However, several intrinsic limitations are involved in the definition, application and reliability of such approach [1].