Room: Track 2
Purpose: DTS has been proposed for image-guided radiotherapy (IGRT) due to its lower dose and shorter scanning time than CBCT. However, due to its small scan angle, DTS has limited volumetric information with severely distorted anatomical structures, which significantly affects its capability for target localization. Conventional deep learning models trained using group patients’ data ignore the inter-patient variabilities, resulting in blurred or inaccurate edges in the augmented DTS especially when scanning angle is less than 90°. In this study, we explored the feasibility to fully restore the volumetric information in DTS by integrating the patient-specific prior information into model training.
Methods: Patient-specific prior CT data were expanded by applying translations, rotations and deformations to mimic various patient on-board positions. Cone-beam projections were simulated from the prior CTs within a 90° scan, and were then used to reconstruct DTS using FDK. A deep learning model was trained on this patient-specific dataset to augment DTS to match with the corresponding ground-truth CT. This patient-specific model was tested for augmenting the DTS acquired on a different day. The augmented DTS was evaluated both qualitatively and quantitatively using root mean squared error (RMSE), peak signal-to-noise ratio (PSNR) and structure similarity (SSIM). An evaluation on tumor localization accuracy was performed to validate the clinical value of the proposed method.
Results: DTS augmented by the proposed method showed accurate and clear edges and excellent agreement with the ground truth CT images with RMSE=0.03, PSNR=30.32, and SSIM=0.92. The augmented DTS demonstrated =0.7mm three-dimensional localization errors compared to the ground truth CT-CBCT registration. Besides, it only takes about 0.9 seconds on GPU for the DTS augmentation.
Conclusion: The proposed method is effective and efficient in restoring volumetric information in DTS reconstructed from only a 90° scan, which significantly boosts the clinical value of using DTS for IGRT applications.
Funding Support, Disclosures, and Conflict of Interest: This work was supported by the National Institutes of Health under Grant No. R01-CA184173 and R01-EB028324.