Document Type
Conference Proceeding
Publication Date
4-30-2022
Publication Title
Radiotherapy and Oncology
Abstract
Purpose or Objective To investigate the feasibility of using an MRI-based synthetic CT image (SCT) generated via a generative adversarial network (GAN) for intensity-modulated proton therapy (IMPT) treatment planning of nasopharyngeal carcinoma (NPC) patients. Materials and Methods T1-weighted MR images and paired CT (PCT) images were obtained from 158 NPC patients with radiotherapy immobilization. Deformable image registration was performed between each MR and PCT image for each patient to create an MR-CT pair. Thirteen pairs were randomly chosen as independent test sets and the remaining 145 pairs (10 for validation and 135 for training) were used to build a conditional GAN model, including a residual-Unet as a generator and a 6-layer convolution neural network as a discriminator. For each test patient, SCT was generated using the generator with the MR image as input. A 4-beam IMPT plan was created and optimized on the corresponding PCT, and the dose matrix was recalculated on the SCT. The dosimetric accuracy was evaluated by using the clinically relevant dose-volume histogram (DVH) parameters and 3D gamma index analysis. Results The mean absolute error between the PCT and SCT images were (89.64+20.54)HU within the body. Figure 1 shows the MR, PCT, SCT, and HU errors for a patient with an average performance of CT number accuracy. The DVH parameters discrepancy between dose matrices calculated on PCT and SCT were (0.13+0.13)%, (0.4+0.44)%, (0.81+0.78)%, (1.25+1.26)%, (1.24+0.78)%, and (1.35+1.1)% for CTV1-D95, CTV2-D95 (involved nodes), Left-parotid mean dose, right-parotid mean dose, brain stem D1, and spinal cord D1, respectively. Figure 2 shows the dose matrices calculated on the PCT and SCT as well as the DVH of critical structures and targets for a patient with an average performance of dosimetric accuracy. The 3%/3mm (10% threshold) gamma passing rate was (97.26+2.35)% within the head and neck region for the 13 test patients.
Volume
170
Issue
Supplement 1
First Page
S1281
Last Page
S1282
Recommended Citation
Chen S, Peng Y, Liu Y, Zhao C, Deng X, Qin A, et al [ Yan D, Stevens C, Deraniyagala R, Ding X.] MRI-based synthetic CT images for IMPT treatment planning of nasopharyngeal carcinoma patients. Radiother Oncol. 2022 Apr;170(Supp 1):S1821-S1282.