(Monday, 7/13/2020) 3:30 PM - 4:30 PM [Eastern Time (GMT-4)]
Room: Track 2
Purpose:
Proton treatment is sensitive to anatomy variation, especially for lung patients. With online CBCT increasingly available in clinic for patient setup, we hypothesize that daily proton treatment dose can be accurately reconstructed from CBCT using deep-learning model for Lung patients.
Methods:
33 pairs of pre-treatment CT and first-fraction CBCT were selected from patients treated with proton therapy in our institution for Lung cancer. Deformable registration was performed to eliminate potential anatomy variation between CT simulation and first fraction. The paired CT-CBCT datasets were then divided into training, validation and testing groups(24, 3 and 6). Two types of GAN models, including one generator with orginal-Unet (Org-GAN) or Residual-Unet (Res-GAN) architecture, and one discriminator using a multi-layer convolutional neural net (CNN), were trained from scratch to predict CT from CBCT, with random deformation as data augmentation. All evaluations were performed on the testing group. The model-predicted synthetic CTs(synCT) were compared to the paired CTs in terms of Mean Absolute Error(MAE). The clinical treatment plans were re-calculated on both synCT and paired CT using a commercial Monte-Carlo algorithm. Clinically relevant DVH parameters and ? passing rates were utilized to quantify the dosimetric discrepancy.
Results:
The model-prediction time is 80ms per 2-D slice. The SynCT MAE was 51.5±7.1 and 49.2±6.7HU with the Org-GAN and Res-GAN model, respectively. Compared to the reference doses on the paired CT, the 3%/3mm and 2%/2mm ?-passing rates of the treatment doses on Res-GAN synCT were 99.5±0.4% and 95.3±1.3 %. The mean clinic-relevant DVH parameters were within 1.0%, except for cord D_1cc(1.9±1.1%), mean dose of esophagus(1.3±0.7%).
Conclusion:
With a relatively small number of training patients, the Res-GAN model can predict synCT via CBCT with clinically acceptable dosimetric accuracy. The proposed method could be utilized to monitor treatment dose online and trigger plan adaptation for patients with undesired anatomy/dosimetric variation.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
