Coded aperture computed tomography

Kerkil Choi, David J. BradyProceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE 2009 · DOI

TL;DR

This paper introduces a novel "Compressive X-ray Tomography" (CXT) system that leverages coded apertures to achieve data compression and improve reconstruction fidelity in X-ray imaging. By unifying the analysis of existing coded aperture systems like CASSI, the authors demonstrate how physical coding can effectively implement the principles of compressed sensing for more efficient X-ray data acquisition.

What Problem Does This Paper Address?

Traditional X-ray tomography, while a powerful imaging technique, often requires a large number of measurements to reconstruct an object accurately. This can lead to long acquisition times and high data storage demands. The paper addresses this by exploring how coded apertures, which modulate X-ray projections, can be used to implement "compressive sensing" (CS). CS theory suggests that if an object has a sparse representation in a particular basis, it can be accurately reconstructed from significantly fewer measurements than traditionally required. The challenge is to practically implement these compressive measurements in X-ray systems, especially when dealing with continuous distributions or when analytical inversion is not straightforward.

Methods

The authors propose a unified mathematical framework for analyzing various X-ray Transform Measurement (XTM) systems that utilize coded apertures. They extend the models for Reference Structure Tomography (RST) and Coded Aperture Snapshot Spectral Imaging (CASSI) to develop a new "Compressive X-ray Tomography" (CXT) model. The core idea is that coded apertures introduce multiplexing and coding into the measurement process, effectively creating "physical random projections." This allows for the acquisition of compressed measurements that can then be used for reconstruction using CS algorithms. The paper also discusses the role of sparse representations and the restricted isometry property (RIP) in ensuring successful reconstruction from these compressed measurements.

Datasets

Not discussed in the paper. The paper focuses on theoretical analysis and numerical simulations using synthetic data.

Key Insights and Ideas

  • Unified Framework for Coded Aperture XTMs: The paper successfully unifies the analysis of different coded aperture systems (RST, CASSI) under a single mathematical model, highlighting their underlying equivalency in terms of measurement principles.
  • Physical Implementation of Compressed Sensing: Coded apertures offer a practical way to implement the random projections crucial for compressed sensing in X-ray imaging, moving beyond purely theoretical random sampling.
  • "Regularizing" Effect of Coding: The coded aperture acts as a form of "regularization" by controlling the correlations between measurement vectors, which is essential for stable reconstruction in underdetermined systems.
  • Impact of Sparse Bases on Reconstruction: The choice of sparse basis (e.g., 2D vs. 3D wavelets or total variation) significantly impacts reconstruction fidelity, with 3D bases often yielding better results for objects with high axial resolution.
  • CASSI vs. CXT Object Characteristics: The paper notes that CASSI objects are spatio-spectral, meaning properties can vary differently along spectral and spatial axes, while CXT objects tend to share similar properties across all three axes, especially with high axial resolution. This difference can influence the effectiveness of 3D sparse bases.

Key Results

The paper presents numerical studies and simulations demonstrating the impact of coding on data efficiency and reconstruction fidelity. While specific quantitative results are presented in figures and tables (e.g., PSNR values), the core finding is that coded aperture systems, particularly the proposed CXT, can achieve significantly improved reconstruction quality (higher PSNR) compared to non-coded or less sophisticated methods, especially when leveraging appropriate sparse bases. For instance, in Figure 3, 3D TV minimization for CASSI achieved a PSNR of 34.3 compared to 31.7 for 2D TV. Figure 4 shows even more dramatic improvements for CXT, with 3D TV minimization achieving PSNRs of 42.1 and 44.4 for high and low axial resolution objects, respectively, compared to much lower values for 2D TV minimization.

Limitations and Future Work

  • Limitations: The paper acknowledges that the exact quantification of how changes in incoherence and sparsity affect reconstruction fidelity is difficult. The study primarily relies on numerical simulations with synthetic data, and real-world implementation challenges are not extensively detailed.
  • Future Work: The authors suggest that further exploration of different coded aperture designs and their impact on measurement incoherence and sparsity would be beneficial. Investigating the trade-offs between acquisition speed, data compression, and reconstruction accuracy in more detail for various applications is also a promising direction. The paper also implies that exploring more advanced CS reconstruction algorithms tailored for these specific coded aperture systems could further enhance performance.

Summary updated

Abstract

Diverse physical measurements can be modeled by X-ray transforms. While X-ray tomography is the canonical example, reference structure tomography (RST) and coded aperture snapshot spectral imaging (CASSI) are examples of physically unrelated but mathematically equivalent sensor systems. Historically, most x-ray transform based systems sample continuous distributions and apply analytical inversion processes. On the other hand, RST and CASSI generate discrete multiplexed measurements implemented with coded apertures. This multiplexing of coded measurements allows for compression of measurements from a compressed sensing perspective. Compressed sensing (CS) is a revelation that if the object has a sparse representation in some basis, then a certain number, but typically much less than what is prescribed by Shannon's sampling rate, of random projections captures enough information for a highly accurate reconstruction of the object. This paper investigates the role of coded apertures in x-ray transform measurement systems (XTMs) in terms of data efficiency and reconstruction fidelity from a CS perspective. To conduct this, we construct a unified analysis using RST and CASSI measurement models. Also, we propose a novel compressive x-ray tomography measurement scheme which also exploits coding and multiplexing, and hence shares the analysis of the other two XTMs. Using this analysis, we perform a qualitative study on how coded apertures can be exploited to implement physical random projections by "regularizing" the measurement systems. Numerical studies and simulation results demonstrate several examples of the impact of coding.
BibTeX
@inproceedings{choi2009coded,
  author = {Kerkil Choi and David J. Brady},
  title = {{Coded aperture computed tomography}},
  booktitle = {Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE},
  year = {2009},
  doi = {10.1117/12.825277},
}
Figure