Design Choices in Splitting-Based Self-Supervised Sparse-View CT Reconstruction

Jul 16, 2026

Upload by Sage

Self-supervised data splitting is used for sparse-view CT reconstruction to train from incomplete measurements without fully sampled ground truth, yet the impact of key design choices is not well understood. A unified framework decomposes splitting-based reconstruction into partitioning strategy, preprocessing, and inference, enabling controlled comparisons and extensions including multi-partition splitting and an alternative inference approach. Experiments on simulated LoDoPaB-CT with independent vs correlated noise and validation on the 2DeteCT dataset identify noise-structure-dependent optimal strategies and show metric sensitivity using LPIPS and HaarPSI.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages21

Reading Time53 min

Introduction

Inverse problems and sparse measurements

Self-supervised splitting for CT reconstruction

Unified framework and research questions

Evaluation metrics and perceptual quality

Design Choices in Splitting-Based Self-Supervised Sparse-View CT Reconstruction Nadja Gruber∗ , Lukas Neumann†, Ander Biguri‡, Gyeongha Hwang§ , Markus Haltmeier¶ , Johannes Schwab ∥ arXiv :2607 . 10898v1 [ cs .CV] 12 Jul 2026 Abstract—Self-supervised data splitting has emerged as a promising paradigm for sparse-view CT reconstruction, enabling training from incomplete measurements without fully sampled ground truth. However, the influence of key design choices, including partitioning strategy, preprocessing, and inference, remains insufficiently understood. In this work, we introduce a unified framework that decomposes splitting-based reconstruction into these three components, enabling controlled comparison of existing methods and two incremental extensions: multi-partition splitting and an alternative inference strategy. Experiments on simulated LoDoPaB-CT data under independent and correlated noise, together with validation on the real-world 2DeteCT dataset, show that the optimal partitioning strategy strongly depends on the measurement noise structure. Lattice-based splitting performs favorably under independent noise, whereas angular masking is more robust under correlated noise and real measured data. Multi-partition splitting consistently improves over pure projection-wise splitting in several settings. Complementary perceptual and structural metrics, including LPIPS and HaarPSI, reveal differences between masking strategies that are less apparent from PSNR and SSIM alone. These results provide practical guidelines for designing self-supervised sparse-view CT reconstruction methods and highlight the limitations of common independence assumptions in realistic imaging environments. Index Terms—Computed tomography, self-supervised learning, sparse-view reconstruction, data partitioning, masking strategies, inverse problems I. INTRODUCTION In imaging modalities such as X-ray computed tomography (CT), cryo-electron microscopy, and photoacoustic imaging, images are reconstructed from indirect and often incomplete measurements. In practice, acquisition constraints such as limited scan time, radiation dose, or physical limitations lead to severely undersampled inverse problems of the form y = N (Ax), (1) where A : X → Y denotes the forward operator, N represents a noise corruption process, x ∈ X is the unknown image, and y ∈ Y are the measured data. Such inverse problems are often ill-posed due to the ill-conditioned nature of A and the presence of noise, meaning additional prior information is required for stable reconstruction. Corresponding authors: [nadja.gruber@uibk.ac.at](nadja.gruber@uibk.ac.at), johannes.schwab@fh[kufstein.ac.at](kufstein.ac.at) ∗ Department of Computer Science, University of Innsbruck, Austria. † Institute of Basic Sciences in Engineering Science, University of Innsbruck, Austria. ‡ Department of Applied Mathematics and Theoretical Physics, University of Cambridge, UK. § Department of Mathematics, Yeungnam University, Gyeongsan, Korea.¶ Department of Mathematics, University of Innsbruck, Austria.∥ University of Applied Sciences Kufstein, Austria. Classical variational methods incorporate handcrafted regularizers such as smoothness or total variation [20], [19], [3],[24] . While theoretically well established, their performance degrades in strongly undersampled regimes and depends heavily on careful parameter selection. More recently, deep learning approaches have achieved state-of-the-art performance by learning image priors directly from data. However, most methods rely on supervised training with paired measurements and ground-truth images, which are often unavailable in practical clinical or industrial imaging settings [2] . To overcome this limitation, self-supervised learning approaches have emerged to exploit structure within the measurements themselves. A prominent class of such methods is based on measurement splitting, where observations are divided into compleme

FAQs

What problem does the document address in sparse-view CT reconstruction?

It studies how design choices in self-supervised data splitting affect sparse-view CT reconstruction when training uses incomplete measurements instead of fully sampled ground truth.

Which components does the proposed unified framework decompose splitting-based reconstruction into?

It decomposes the method into three components: partitioning, preprocessing of masked measurements, and inference strategy.

How does measurement noise structure influence the best splitting strategy?

The results show that optimal partitioning depends on whether noise is independent or correlated: lattice-based splitting performs better under independent noise, while angular masking is more robust under correlated noise and real measured data.

Jul 16, 2026

Upload by Sage