R. Zhang1,2, and J. Zhu1; 1Shandong Cancer Hospital and Institute, Shandong First Medical University and Shandong Academy of Medical Sciences, Jinan, China, 2Linyi University, Linyi, China
Purpose/Objective(s): To investigate the accuracy of proton dose calculation using synthetic CT (sCT) images generated by a registration generative adversarial network (RegGAN) model from cone-beam CT (CBCT) images. Dosimetric differences between sCT-based and planning CT (pCT)-based proton radiotherapy plans were compared to explore the feasibility of sCT in brain tumor proton adaptive radiotherapy. Materials/
Methods: The study collected datasets from 10 patients with cranial tumors, including pCT images acquired on a technology company SOMATOM Confidence large-bore simulation positioning system and CBCT images obtained in real-time before treatment in the accelerator room using the technology company ProBeam proton accelerator with onboard kilovoltage cone-beam CT. The study focused on a set of CBCT images scanned before the first treatment fraction. The RegGAN network was employed to transform CBCT into sCT images. To visualize the Hounsfield Unit (HU) differences between sCT and pCT, a pseudo color map reflecting their HU differences was generated by layer-by-layer silhouette according to HU values. For dosimetric comparison, sCT was imported into Eclipse TPS, and the relevant structural contours and treatment plans from pCT were rigidly transferred to sCT images. The corresponding CT calibration curves were assigned to ensure consistency, and the proton convolution superposition (PCS) algorithm in Eclipse TPS was used for forward dose calculation on sCT with pCT treatment plan parameters. After all plan dose calculations were completed and passed the prescription audit, the two plans were evaluated and compared using dose-volume histogram (DVH) analysis. Results: The sCT images generated via the RegGAN method showed improved image quality compared to the original CBCT. The HU difference map indicated that significant HU discrepancies were mainly concentrated in bone-containing regions (cranial and nasal bones), with the largest difference in the skull averaging 84 HU, while soft tissues showed minimal disparity. The DVHs of sCT and pCT plans for the selected 10 cranial tumor patients matched well. The Dmean for CTV in sCT and pCT plans were 103.940±0.693% and 104.120±0.624%, respectively, with a maximum error of 0.8% and an average of 0.2%. The conformity index (CI) had a maximum error of 4.1% and an average of 1.7%. Conclusion: Preliminary studies indicate that sCT images generated by the RegGAN model have high image quality and HU values closer to those of pCT, resulting in similar dosimetry factors to pCT plans within acceptable errors. Therefore, sCT images generated by the RegGAN method from CBCT can be used for proton adaptive radiotherapy in cranial tumors. Keywords: Cone-beam CT; Synthetic CT; RegGAN; Proton adaptive radiotherapy
