(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: On-board cone-beam CT (CBCT) is commonly used during radiation therapy for patient positioning and other purposes. Four-dimensional (4D) CBCT is desirable for motion-resolved imaging of moving tumors, e.g., for lung, pancreatic and liver cancer patients. However, high-quality 4D CBCT image reconstruction remains unsolved as a sparse-data problem.
Methods: Here we develop a new method, namely 4D-AirNet, that synergizes analytical method, iterative method, and deep learning for high-quality 4D CBCT image reconstruction. 4D-AirNet is an unrolling method using the optimization framework of fused analytical and iterative reconstruction (AIR), which is based on proximal forward-backward splitting (PFBS). Three different strategies are developed for 4D-AirNet: random-phase (RP), prior-guided (PG), and all-phase (AP). RP-AirNet and PG-AirNet utilize phase-by-phase training and reconstruction, while PG-AirNet also uses a prior image reconstructed with all ten-phase projection data. Dense connectivity is built into 4D-AirNet networks for improved reconstruction quality. In contrast, AP-AirNet trains and reconstructs all phases simultaneously. In addition, the joint regularization method of DL and conventional spatiotemporal total variation (TV) is investigated.
Results: 4D-AirNet methods were evaluated in comparison with state-of-art iterative (TV) and deep learning (LEARN) methods, using simulated 4D CBCT scans from a lung dataset with various sparse-data levels. The reconstruction results suggest 4D-AirNet methods outperform TV and LEARN, and AP-AirNet provides the best reconstruction quality overall.
Conclusions: 4D-AirNet methods have been proposed for high-quality 4D CBCT image reconstruction, including RP-AirNet, PG-AirNet and AP-AirNet. RP-AirNet and PG-AirNet train the networks phase-by-phase, while AP-AirNet trains the networks for all phases simultaneously. PG-AirNet implicitly uses all phases during each training epoch by including a prior image, while AP-AirNet explicitly uses all phases during each training epoch by all-phase training.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
