Vorain-Gan: Volumetric Attention Gan Trained with Calibrated MVCT for Kvct Synthesis In Head and Neck Cancer of Tomotherapy
Abstract
Purpose
During radiotherapy, anatomical changes may impact the suitability of the initial treatment plan, necessitating the application of adaptive radiation therapy (ART). The MVCT serves as the patient positioning image, enabling dose reconstruction. However, its limited image quality limit its application in ART. This study was to develop an image synthesis network, VoRAIN GAN, that further improved the image quality of synthetic kVCT (skVCT) based on prior work.
Methods
MVCT and corresponding kVCT data from 63 patients with head and neck cancer were collected. 41 patients were used for training, while 22 were reserved for testing. This study employs a CycleGAN network incorporating attention gates, residual blocks, and volume control loss, named “Volumetric Residual Attention-Integrated GAN (VoRAIN-GAN)” . The MVCT images were calibrated using the CT values-electron density (CT-ED) curve of MVCT and kVCT. The VoRAIN-GAN was trained using calibrated MVCT. The evaluation indicators include mean absolute error (MAE), structural similarity index measurement (SSIM), peak signal-to-noise ratio (PSNR), and gamma passing rate.
Results
The MAE, PSNR, and SSIM of MVCT were respectively 102±14.9 HU, 27.3±1.4 dB, and 0.917±0.018, and those of skVCT generated by the previously published model were 67.9±13.9 HU, 31.4±2.0 dB, and 0.960±0.02, while those of skVCT in this study were 59.6±6.0 HU, 32.7±1.2 dB, and 0.965±0.017, respectively. The gamma passing rates improved from 96.39 ± 2.13% to 97.13 ± 1.81% (1 mm/ 1%), 98.63 ± 0.95% to 98.86 ± 0.81% (2 mm/ 2%), respectively.
Conclusion
This study developed a Cal VoRAIN-GAN to generate MVCT-based skVCT. The proposed method effectively enhances image quality and improves dose calculation accuracy, as demonstrated in the original work, indicating potential to improve treatment accuracy and enable ART implementation. Funding: This work was supported by the National Natural Science Foundation of China (No. 12375359) and CAMS Innovation Fund for Medical Sciences (CIFMS, 2024-I2M-C&T-B-076).