l tie^1,2*, C Xie², X Dai², W Zhao¹, G Zhang¹, B Qu², S Xu³, (1) Beihang University, School of Physics, Beijing, China, (2) The First Medical Center of PLA General Hospital, Department of Radiation Oncology, Beijing, China.,(3) National Cancer Center/Cancer Hospital- Chinese Academy of Medical Sciences and Peking Union Medical College, Department of Radiation Oncology, Beijing, China

PO-GePV-I-22 (Sunday, 7/10/2022)   [Eastern Time (GMT-4)]

ePoster Forums

Purpose: Megavoltage computed tomography (MVCT) offers the possibility of adaptive helical tomotherapy. However, MVCT has low contrast and noises, making it challenging to delineate. This study developed a deep-learning-based approach to generate high-quality synthetic kilovoltage computed tomography (skVCT) from MVCT and meet dose requirements.

Methods: Data from 31 head and neck cancer patients were collected, 25 (2995 slices) used for training and 6 (698 slices) for testing. A cycle generative adversarial network (cycleGAN) based on attention gate and residual blocks generated MVCT-based skVCT. For the six patients, kVCT-based contours and treatment plans were directly transferred to skVCT images and electron density profile-corrected MVCT images to calculate dose. The quantitative indices included the mean absolute error (MAE), Mean error (ME), root-mean-square error (RMSE), structural similarity index measure (SSIM), peak signal-to-noise ratio (PSNR), gamma passing rate and dose-volume-histogram (DVH) parameters (Dmax, Dmean, Dmin, D95).

Results: The MAE, ME, RMSE, PSNR, and SSIM of MVCT were 82.73±12.05 HU, 19.02±10.48 HU, 136.43±21.63 HU, 32.99±1.54 dB, and 98.78±0.56, and that of skVCT were improved to 44.04±6.33 HU, -0.16±10.88 HU, 76.03±11.88 HU, 40.30±2.65 dB, and 99.42±0.33. Image quality and contrast were improved, and the noises were reduced. The gamma passing rates improved from 98.74%±1.38% to 99.80%±0.18% (2 mm/2%), 94.6%±4.39%% to 98.34%±0.92% (1 mm/2%), and 85.79%±7.95% to 93.97%±3.02% (0.5 mm/2%). The overall dose distribution of skVCT is similar to that of kVCT. For DVH parameters, no significant differences (p > 0.05) were observed between kVCT and skVCT for PTV and OAR, while significant differences (p < 0.05) were found between kVCT and MVCT.

Conclusion: With training on a small data set (2995 slices), the generated skVCT improved the image quality and was superior to MVCT dose calculation accuracy. MVCT-based skVCT has the potential to increase treatment accuracy and offers the possibility of adaptive helical tomotherapy.

Keywords

Not Applicable / None Entered.

Taxonomy

IM/TH- Image Analysis Skills (broad expertise across imaging modalities): image synthesis/simulation and augmentation

Contact Email