Main Session
Sep
30
PQA 09 - Hematologic Malignancies, Health Services Research, Digital Health Innovation and Informatics
3657 - Adjacent-Double Projections-Based CBCT Reconstruction for Adaptive Radiotherapy
Presenter(s)
Danyang Li, PhD - Sun Yat-Sen University Cancer Center, Guang Dong Province, Guangdong
D. Li, X. Huang, Y. Sun, and L. Lin; State Key Laboratory of Oncology in South China, Guangdong Key Laboratory of Nasopharyngeal Carcinoma Diagnosis and Therapy, Guangdong Provincial Clinical Research Center for Cancer, Department of Radiation Oncology, Sun Yat-sen University Cancer Center, Guangzhou, Guangdong, China
Purpose/Objective(s):
Cone-beam computed tomography (CBCT) plays a crucial role in adaptive radiotherapy (ART), enabling the detection and management of anatomical changes during treatment. However, the current on-board CBCT systems are limited by slow scan times, typically taking around one minute to complete a full acquisition, which hinders real-time application in ART. To address this bottleneck, we introduced a novel deep-learning framework, referred to as ADP-CBCT, which leverages Adjacent-Double Projections to accelerate CBCT image acquisition and reconstruction. This innovative approach not only streamlines the ART process but also reduces patient radiation exposure.Materials/Methods:
The DAVP-CBCT model is composed of three key modules: the encoder, the transfer module, and the generator. First, the encoder extracts semantic features from the adjacent-double projections, capturing the critical information necessary for CBCT image reconstruction. Next, the transfer module converts these features from the projection domain into the image domain, performing a flattening and transformation process followed by reshaping. Finally, the generator synthesizes the complete volumetric CBCT image, delivering the final reconstructed output.Results:
The model was evaluated using a total dataset of 300 NPC patients, divided into 210 for training, 30 for validation, and 60 for testing datasets. Quantitative performance metrics included mean absolute error (MAE), structural similarity index measure (SSIM), and peak signal-to-noise ratio (PSNR). On the testing dataset, the DAVP-CBCT model achieved an average MAE of 35±10.1 HU (mean±std), an SSIM of 0.9625±0.0115, and a PSNR of 38.68±1.35 dB, demonstrating its high accuracy and robustness.Conclusion:
In this study, we developed a deep learning-based model for CBCT reconstruction aimed at accelerating the imaging process. Quantitative evaluations confirmed that the model successfully produces high-quality 3D volumetric CBCT reconstructions. Notably, the model requires only adjacent-double projections, making it highly compatible with existing image-guided radiotherapy workflows. This integration capability highlights its potential to enable real-time imaging in ART.