Cycle-GAN-based image correction for chest cone-beam CT and accuracy of dose calculation(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2022年第7期
- Page:
- 811-816
- Research Field:
- 医学放射物理
- Publishing date:
Info
- Title:
- Cycle-GAN-based image correction for chest cone-beam CT and accuracy of dose calculation
- Author(s):
- WU Xianxiang1; NIU Zhenyang2; CAI Hanfei1; FANG Meifang1; LI Wei1; XU Lu1; CUI Zhen1
- 1. Department of Radiation Oncology, the First Af?iated Hospital of Bengbu Medical College, Bengbu 233004, China 2. Department of Radiation Oncology, No.901 Hospital of PLA, Hefei 230031, China
- Keywords:
- Keywords: cycle-GAN cone-beam computed tomography image correction dose calculation adaptive radiotherapy
- PACS:
- R318
- DOI:
- DOI:10.3969/j.issn.1005-202X.2022.07.004
- Abstract:
- Abstract: Objective To establish a correction model for chest cone-beam CT (CBCT) using cycle-GAN algorithm for exploring the feasibility of using the model to improve CBCT image quality, and evaluating the accuracy of the corrected CBCT (CCBCT) for dose calculation. Methods The registered CBCT and positioning CT were collected from 70 patients with esophageal cancer and lung cancer. Sixty of them were randomly selected as training set to train cycle-GAN and generate the correction model for CBCT, while the remaining 10 cases were used as test set. The average absolute error of Hounsfield unit, peak signal-to-noise ratio, and normalized cross-correlation between CBCT or CCBCT and positioning CT were analyzed. CCBCT Plan was generated by transplanting the original intensity-modulated radiotherapy plan (CT Plan) to CCBCT. Taking the dose distribution of CT Plan as a reference, the gamma analysis on three-dimensional dose verification in CCBCT Plan was performed. Results Scattering artifacts were significantly reduced after CBCT was corrected. The average absolute error of Hounsfield unit was reduced by (52.74±6.47)%, and the peak signal-to-noise ratio and normalized cross-correlation were increased by (7.95±3.56)% and (1.68±3.38)%, respectively, with statistically significant differences (t=18.47, -7.31, -6.77 P[<]0.05). Under the criteria of 2 mm/2%, 2 mm/1%, and 1 mm/1%, the average gamma passing rates of three-dimensional dose verification in CCBCT Plan were (99.16±0.34)%, (98.01±0.72)%, and (93.95±1.62)%, respectively. Conclusion Using the cycle-GAN-based correction model for CBCT to improve the CBCT image quality is proved to be feasible. The corrected chest CBCT can be used for radiation dose calculation, laying a foundation for CBCT to be used for adaptive radiation dose calculation.
Last Update: 2022-07-15