Room: Stars at Night Ballroom 2-3
Purpose: To develop a generative adversarial network (GAN) method to produce synthetic CT (sCT) images for MRI-only proton treatment planning with accurate geometric and Hounsfield Units (HU) information. Dosimetric evaluation was performed using an independent Monte Carlo (MC) dose engine.
Methods: A modified conditional GAN model was designed to learn a direct mapping function to convert a MRI slice to its corresponding CT slice. The model was trained with T1-weigthed MRI slices and its corresponding CT slices from 66 brain tumor patients, and tested in 11 patients. Intensity modulated proton therapy (IMPT) plans were generated on the CT with RayStation software and recomputed on the sCT using a fast MC dose engine, for all test patients. The dosimetric accuracy and plan robustness was analyzed with relevant DVH metrics.
Results: The average of mean absolute error (MAE) Â± SD of the HU values in the generated sCT for all test patients was equal to 47.2 Â± 11.0 HU. The time to generate a full three-dimensional sCT for each patient was about 1s. The dosimetric evaluation showed an excellent agreement between the doses computed on the CT and sCT, with most DVH metrics presenting a mean absolute difference below 0.5% (0.3 Gy) of the prescription dose (60 Gy) for the target and below 2% (1.2Gy) for the considered organs.
Conclusion: A modified C-GAN model was developed, and shown to be able to produce highly accurate sCT from conventional, single-sequence MRI images in nearly real time. Quantitative results also showed that the proposed method can generate sCT images with improved accuracy and faster computation speed compared to conventional methods. This work demonstrated the feasibility of using sCT generated with a deep learning method based on GANs, for MRI-only treatment planning in intensity modulated proton therapy.
Not Applicable / None Entered.