Search papers, labs, and topics across Lattice.
This paper introduces a Gaussian-Based Shift-Variant filtered backprojection (FBP) neural network tailored for efficient reconstruction in non-circular trajectory cone beam computed tomography. By integrating a trainable 2D Gaussian model into the filtering process, the approach achieves a remarkable 99% reduction in trainable parameters with minimal loss in reconstruction quality. Additionally, the training time for each trajectory is cut to a quarter of the original, enhancing both the model's practicality and effectiveness for real-world applications.
Achieving a 99% reduction in parameters without significant quality loss could revolutionize the efficiency of cone beam CT reconstructions.
This paper proposes a Gaussian-Based Shift-Variant filtered backprojection (FBP) neural network, which is designed for the efficient reconstruction of non-circular trajectory cone beam computed tomography. The traditional differentiable shift-variant FBP model consists of a filtering component and a backprojection process. The filtering component includes operations such as weightings, differentiations, a 2D Radon transform, and a 2D backprojection. The proposed methods build on this framework by introducing a trainable 2D Gaussian model to represent the trajectory-related part in the filtering process, achieving a substantial reduction in the number of trainable parameters. Experimental results demonstrate that the proposed model reduces the parameter count by 99%, while only sacrificing a slight amount of reconstruction quality. Furthermore, the training time for each trajectory is reduced to one-fourth of the original, significantly accelerating convergence. These enhancements demonstrate a considerable augmentation in the model's practicality and effectiveness, making it a valuable asset for real-world applications.
