A Sparse CBCT 3D Reconstruction Framework with Enhanced Global Information

CBCT imaging plays a critical role in clinical applications, particularly in radiotherapy. Due to safety concerns arising from the cumulative radiation dose caused by extensive X-ray exposure, sparse-view and low-dose CBCT imaging have been identified as important approaches to address this issue. However, sparse-view and low-dose CBCT imaging suffer from poor image quality and severe artifacts. Recent years have witnessed the emergence of numerous deep learning methods aimed at tackling this problem. Yet, many of these methods neglect the global information in the reconstructed 3D images when designing network architectures. Addressing this limitation is of great significance. Building upon the deep intensity field approach, we propose a sparse-view CBCT reconstruction framework that prioritizes global information by incorporating an attention mechanism and the S3IM module. Experiments demonstrate that our method achieves superior performance on the CQ500 dataset compared to existing models.