Knowledge-Based Deep Residual U-Net for Synthetic CT Generation Using a Single MR Volume for Frameless Radiosurgery

Abstract

AbstractBackground: Stereotactic radiosurgery (SRS) requires MRI imaging for accurate target delineation. Synthetic CT (sCT) generated from MRI offers a streamlined MR-only solution without additional CT simulation. However, challenges remain in sCT accuracy for SRS planning in practical settings, especially when considering immobilization devices. Purpose: To develop a knowledge-based deep model for sCT generation from a single MR volume in LINAC-based frameless SRS, enabling an MR-only workflow without the need for extra CT simulation. Methods: A total of 139 patients were included in the study, with 120 used for training and 19 for testing. A Deep Residual U-Net (DRU) was developed to generate sCT from patient-specific high-resolution T1+C MR volume, complemented by a healthy brain CT volume from the Visible Human Project, which provides CT-specific anatomical knowledge. To simulate treatment conditions, a template immobilization mask was deformed to align with the patient-specific sCT anatomy, creating a full sCTF volume. Four metrics, including PSNR, SSIM, RMSE, and MAE, were derived to evaluate the Hounsfield units (HU) accuracy of sCT compared to the ground-truth CT without immobilization masks. Multi-target SRS plans developed with the VMAT technique were recalculated within sCTF volumes to produce simulated dose distributions. These were compared with clinical plan dose distributions using the mean dose difference in the planning target volume (PTV) and gamma index evaluation. Results: In the test set, the generated sCT achieved a PSNR of 75.40 ± 3.58 dB, SSIM of 0.99 ± 0.01, RMSE of 11.88 ± 5.82 HU, and MAE of 1.38 ± 0.81 HU for brain tissues. When comparing sCTF dose calculation results against the original plan, gamma index passing rates were 95.77 ± 4.17% for the entire volume and 84.36 ± 14.95% within PTVs, using 3%/1 mm/15% threshold criteria. The PTV mean dose difference averaged -2.32 ± 1.48%, with most discrepancies below -5.0%. Conclusion: This study successfully demonstrated the generation and validation of sCT images from single-modality MRI using a knowledge-based deep model. The results confirm that single-modality MRI effectively supports frameless SRS and integrates seamlessly into current clinical workflows.