Abstract
X-ray computed tomography (CT) is a powerful tool for noninvasive imaging of time-varying objects. In the past, methods have been proposed to reconstruct images from continuously changing objects. For discretely or structurally changing objects, however, such methods fail to reconstruct high quality images, mainly because assumptions about continuity are no longer valid. In this paper, we propose a method to reconstruct structurally changing objects. Starting from the observation that there exist regions within the scanned object that remain unchanged over time, we introduce an iterative optimization routine that can automatically determine these regions and incorporate this knowledge in an algebraic reconstruction method. The proposed algorithm was validated on simulation data and experimental μCT data, illustrating its capability to reconstruct structurally changing objects more accurately in comparison to current techniques.
Introduction
Standard X-ray Computed Tomography (CT) relies on a fundamental assumption: the object must remain perfectly still while the X-ray source rotates around it. If the object moves or changes during the scan, the resulting image suffers from "motion blur" and streak artifacts.
Existing solutions generally fall into two categories:
- Deformation-based methods: These model the object as a continuous "stretching" mass. They are effective for smooth movements like a beating heart but fail when changes are structural or discontinuous, such as a new crack appearing or a void filling with liquid.
- Subset-based methods: These take small "snapshots" using only a few projection angles. While this captures temporal changes, the lack of data leads to extremely noisy, low-quality images.
This paper identifies a middle ground: in many industrial and biological scenarios, large parts of the object remain stationary while only a specific sub-region undergoes change. Current methods do not effectively exploit this "static background" redundancy.
Methods: The rSIRT Algorithm
The core contribution is the Region-Based SIRT (rSIRT) algorithm, which incorporates an automated method to identify changing regions.
Unlike the standard Simultaneous Iterative Reconstruction Technique (SIRT), rSIRT treats the image as two distinct zones:
- Stationary Region ($I_S$): This region uses all available projection data from the entire scan duration to ensure a high signal-to-noise ratio.
- Variable Region ($I_V$): This region only uses a small "window" of projections centered around a specific time point to capture the dynamic state accurately.
Automatic Region Estimation
The algorithm does not require manual input to locate changes. Instead, it uses a B-spline closed curve model to define the boundary of the changing area. An optimization routine, utilizing the Levenberg-Marquardt method, minimizes Region Inconsistency (rI). This process "hunts" for the best boundary by minimizing the difference between actual X-ray measurements and simulated projections of the reconstructed model.
Datasets and Validation
The authors validated their approach using both simulated and real-world data:
- Numerical Simulations: Four distinct "phantoms" were used, including a modified Shepp-Logan phantom simulating liquid flow and a material block where a crack suddenly forms.
- Experimental μCT Data: A physical Plexiglas resolution phantom was scanned twice—once dry and once with holes filled with water. These datasets were mathematically combined to mimic a hole filling mid-scan.
Key Results
The rSIRT method demonstrated significant improvements in both quantitative and qualitative metrics:
- Superior Accuracy: In the "crack formation" simulation, rSIRT achieved a Root Mean Square Error (RMSE) of 0.28, significantly outperforming standard SIRT (0.36) and conventional subset-based methods (0.36).
- Visual Clarity: Qualitatively, rSIRT produced sharp edges in the changing regions where other methods showed heavy blurring or "ghosting" artifacts.
- Temporal Resolution: The method allows for much shorter acquisition windows for the dynamic parts, effectively increasing the "frame rate" of the CT scan without sacrificing the image quality of the background.
Key Insights
- Exploiting Redundancy: By "locking in" the static parts using the full dataset, the algorithm focuses its limited temporal data purely on the changing pixels.
- Confidence Weighting: A specialized weighting matrix ($R_\alpha^\lambda$) gives higher "trust" to X-ray rays that pass mostly through stationary regions.
- Structural Focus: This shifts the focus from simple motion to structural changes, which is critical for non-destructive testing in engineering and soil science.
Limitations and Future Work
While effective, the optimization process to find the region boundary is computationally intensive, taking 2 to 4 hours in the study. Future work involves GPU parallelization to enable real-time use. Additionally, the current B-spline model assumes a single, continuous changing region; future iterations could support multiple disconnected areas. Finally, the model assumes the "static" part is perfectly still, which may not hold in environments with vibrations or thermal expansion.
Summary updated