A Deep Learning Method to Improve the Quality for High-Speed Imaging in a 1.5 T MRI Radiotherapy System
J Zhu*, X Chen, B Yang, R Wei, S Qin, Z Yang, Z Hu, J Dai, K Men, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences, Beijing, 11CN,
Presentations
SU-H330-IePD-F5-3 (Sunday, 7/10/2022) 3:30 PM - 4:00 PM [Eastern Time (GMT-4)]
Exhibit Hall | Forum 5
Purpose: MRI represents a new imaging method for online adaptive radiotherapy. Images acquired with a fast-scanning sequence can reduce time, but the inferior image quality may not meet the clinical requirements. The purpose of this study was to improve image quality for high-speed imaging using a deep learning method. We also evaluated its benefit on image registration.
Methods: Thirty fractions of 1.5T MR images acquired with an MR-linac were enrolled. The data included low-speed, high-quality (LSHQ) and high-speed, low-quality (HSLQ) MR images. We proposed a CycleGAN to learn the mapping between the HSLQ and LSHQ images and then generate synthetic LSHQ (synLSHQ) images from the HSLQ images. A five-fold cross-validation was employed to test the CycleGAN model. The normalized mean absolute error (nMAE), peak signal-to-noise ratio (PSNR), structural similarity index measurement (SSIM), and edge keeping index (EKI) were calculated to determine the image quality. The Jacobian determinant value (JDV) and the interested structure errors were used to analyze deformable registration.
Results: The proposed model can generate synthetic MRI images visually comparable to that of low-speed, high-quality sequences which are used in clinical practice. This method (combining with HSLQ and synLSHQ) saved 59.8% more time for a single case. Meanwhile, the quantitative metrics of the image quality were better than that of HSLQ images, with an improvement of 57%, 4.6%, 28.1%, and 5.6% for nMAE, SSIM, PSNR, and EKI, respectively. Furthermore, the generated synLSHQ also enhanced the registration accuracy with a superior mean Jacobian determinant value (2.3%) and preferable interested structures errors.
Conclusion: The proposed method can generate high-quality images from high-speed scanning sequences. Therefore, it can shorten the overall treatment time while ensuring the accuracy of radiotherapy.
Funding Support, Disclosures, and Conflict of Interest: This study has received funding by the National Natural Science Foundation of China (11975313, 12005302).
Keywords
MRI, Prostate Therapy, Image Artifacts
Taxonomy
Not Applicable / None Entered.
Contact Email