GB-SVFBP: Gaussian-Based Shift-Variant FBP neural network

2026-07-13

AI Summary

This work addresses the high parameter count and low training efficiency of conventional differentiable shift-variant filtered back-projection (FBP) models in cone-beam CT reconstruction with non-circular trajectories. To overcome these limitations, the authors propose a lightweight modeling approach based on a learnable two-dimensional Gaussian kernel. Within the shift-variant FBP framework, this method explicitly represents trajectory-dependent filtering responses using a parametric 2D Gaussian function, replacing the original complex filter components. The proposed model reduces the number of trainable parameters by 99% and shortens single-trajectory training time to one-quarter of the original, while incurring only a marginal degradation in reconstruction quality. Consequently, the approach substantially enhances training efficiency and deployment feasibility without compromising image reconstruction performance.

Abstract

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.

Problem

• cone beam computed tomography
• non-circular trajectory
• filtered backprojection
• shift-variant reconstruction
• parameter efficiency

Innovation

• Gaussian-based
• Shift-variant FBP
• Cone beam CT
• Parameter reduction
• Non-circular trajectory