Adaptive Radiotherapy Via Deep Learning-Based Quantitative Cone-Beam CT Imaging
L Wan1*, H Wu2, S Xu3, B Sun4, W Zhao5, (1) Beihang University, Beijing, BJ, CN (2) Beijing Cancer Hospital, Beijing, BJ, CN, (3) National Cancer Center/Cancer Hospital- Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, BJ, CN, (4) Beihang University,Beijing, BJ, CN(5) Beihang University, Beijing, BJ, CN
Presentations
TU-D1000-IePD-F2-5 (Tuesday, 7/12/2022) 10:00 AM - 10:30 AM [Eastern Time (GMT-4)]
Exhibit Hall | Forum 2
Purpose: Adaptive radiotherapy (ART) can address substantial anatomy changes (such as tumor shrinkage or growth) throughout treatment and is therefore of great interest to improve patient outcomes. ART relies on up-to-date quantitative anatomical images which are not usually available in routine practice. Daily cone-beam CT (CBCT) is able to provide up-to-date anatomy information but suffered from insufficient accuracy. Here we perform ART to greatly enhance patient dose conformity by using CBCT-based quantitative synthetic CT (sCT) images.
Methods: To provide quantitative images for adaptive treatment, the CBCT images are firstly resampled and rigidly registered with the planning CT (pCT) images. The resulting images together with pCT images are then employed to train a cycle-consistent adversarial network using a hybrid mean absolute error (MAE) and perceptual loss function. The predicted quantitative CT images are finally subject to dose calculation and replan treatment with physician-approved automatic delineated contours. Fourteen patients who received prostate volumetric modulated arc therapy (VMAT) are used to evaluate the ART method.
Results: The MAE within the region-of-interests (muscle, fat, bladder) between sCT and pCT were 29.0, 26.5, and 25.1 HU, respectively. Taking the calculated results of pCT doses as the standard, for PTV, the gamma pass rates of sCT doses at 1mm/1%, 2mm/2%, and 3mm/3% were 80.6%, 98.3%, and 99.9%, respectively. According to the dose calculation after re-contouring of the bladder using sCT, there were two patients whose bladder dose V40>50%. For these two patients, we performed rapid replanning to reduce the bladder dose to V36<50%,V33<50% under the condition that the PTV met the prescribed dose.
Conclusion: We propose a method to perform adaptive treatment using deep learning synthesized quantitative CT images. This strategy can mitigate the impact of anatomical changes and significantly improve the dose conformity to the target while sparing the normal tissue.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by the National Natural Science Foundation of China (No. 12175012)
Keywords
Prostate Therapy, Cone-beam CT, Dose
Taxonomy
TH- External Beam- Photons: adaptive therapy
Contact Email