Improving the Generalisation of Learned Reconstruction Frameworks

2025-11-16

AI Summary

Poor generalizability of data-driven methods in X-ray CT reconstruction—particularly CNNs, which neglect the intrinsic line-manifold structure of sinograms and over-rely on grid-based convolutions, leading to parameter redundancy and weak cross-geometry generalization—is addressed in this work. We propose the first graph-grid hybrid neural network framework, explicitly modeling projection geometry via a graph structure and pioneering the synergistic integration of graph neural networks (GNNs) with CNNs for sinogram-domain reconstruction. This design endows the model with strong generalization to unseen sampling geometries. Experiments demonstrate significant improvements over pure CNNs in SSIM (+0.025) and PSNR (+1.8 dB), ~40% reduction in trainable parameters, substantially lower training time and memory consumption, and effective transfer from full-sampling training to sparse-view testing.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Reasoning under Uncertainty: Graphical Models

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Ensuring proper generalization is a critical challenge in applying data-driven methods for solving inverse problems in imaging, as neural networks reconstructing an image must perform well across varied datasets and acquisition geometries. In X-ray Computed Tomography (CT), convolutional neural networks (CNNs) are widely used to filter the projection data but are ill-suited for this task as they apply grid-based convolutions to the sinogram, which inherently lies on a line manifold, not a regular grid. The CNNs, unaware of the geometry, are implicitly tied to it and require an excessive amount of parameters as they must infer the relations between measurements from the data rather than from prior information. The contribution of this paper is twofold. First, we introduce a graph data structure to represent CT acquisition geometries and tomographic data, providing a detailed explanation of the graph's structure for circular, cone-beam geometries. Second, we propose GLM, a hybrid neural network architecture that leverages both graph and grid convolutions to process tomographic data. We demonstrate that GLM outperforms CNNs when performance is quantified in terms of structural similarity and peak signal-to-noise ratio, despite the fact that GLM uses only a fraction of the trainable parameters. Compared to CNNs, GLM also requires significantly less training time and memory, and its memory requirements scale better. Crucially, GLM demonstrates robust generalization to unseen variations in the acquisition geometry, like when training only on fully sampled CT data and then testing on sparse-view CT data.

Problem

• CNNs struggle with CT reconstruction generalization across varied datasets
• Grid-based convolutions are ill-suited for sinogram data on line manifolds
• Existing methods require excessive parameters and lack geometry awareness

Innovation

• Graph data structure represents CT acquisition geometries
• Hybrid network uses graph and grid convolutions
• GLM improves generalization with fewer parameters