TL;DR
This paper evaluates the Dual-Energy Iterative Reconstruction Algorithm (DIRA) for accurate CT number determination using real-world dual-energy CT (DECT) data. DIRA successfully reduces beam-hardening artifacts and provides CT numbers that closely match tabulated reference values, demonstrating its potential for quantitative DECT applications like radiation treatment planning.
What Problem Does This Paper Address?
Accurate quantitative measurements in CT, particularly CT numbers (Hounsfield Units or HU), are crucial for various medical applications, including radiation treatment planning. Dual-energy CT (DECT) offers the potential for improved accuracy by acquiring data at two different X-ray energy levels, which can help mitigate artifacts like beam-hardening. Beam-hardening occurs because lower-energy photons are preferentially absorbed, leading to inaccurate attenuation values, especially in heterogeneous materials. While previous simulations showed DIRA's promise in generating artifact-free, monoenergetic images with accurate CT numbers, its performance on real experimental data was not yet established. This paper bridges that gap by testing DIRA on phantom data acquired with a clinical DECT scanner.
Methods
The core of this work is the evaluation of the DIRA algorithm, which is an iterative reconstruction method that incorporates material decomposition within its iterative loop. This approach aims to generate monoenergetic images that are free from beam-hardening artifacts. For experimental data, several adaptations were necessary compared to simulation:
- Inverse Water Beam-Hardening Correction (WBHC): The raw data had already undergone a standard WBHC, which was not ideal for DIRA. An inverse WBHC was developed to restore the original, uncorrected projection data.
- Patient Table Removal: To avoid artifacts from the patient table, it was either positioned at the end of the scanner to allow the phantom to "float" or a reference scan of the table alone was subtracted.
- Bow-tie Filter Modeling: The X-ray beam's spectrum is altered by the bow-tie filter, which is thinnest at the center. DIRA was adapted to model and account for this angular dependence of the spectrum.
- Material Decomposition: DIRA utilizes material decomposition (2-material or 3-material) to represent voxels as mixtures of base materials. For the phantoms used, a water-bone basis was sufficient.
- Iterative Refinement: The algorithm iteratively refines the image reconstruction by comparing forward-projected data with measured projections, updating the material decomposition and attenuation maps until convergence.
Datasets
The study evaluated DIRA using two phantoms:
- Cylindrical Polymethyl Methacrylate (PMMA) Phantom: This phantom contained four cylindrical rod inserts made of aluminum, two of polytetrafluoroethylene (PTFE), and one of low-density polyethylene (LDPE). This allowed for testing material-specific accuracy and beam-hardening effects between dissimilar materials.
- Anthropomorphic Head Phantom (CIRS model 731-HN): This phantom has a specified material composition mimicking human tissues (brain, soft tissue, bone, etc.), providing a more clinically relevant evaluation.
The reference CT numbers for comparison were obtained from the EPDL97 library.
Key Insights and Ideas
- Experimental Validation of DIRA: The study successfully demonstrates that DIRA, when adapted for real-world DECT data, can achieve accurate CT numbers and effectively suppress beam-hardening artifacts. This moves DIRA from a promising simulation-based algorithm to a viable tool for clinical applications.
- Importance of Pre-processing Adaptations: The need for an inverse WBHC and modeling of the bow-tie filter highlights the critical role of accurate pre-processing and system modeling when applying reconstruction algorithms to experimental data.
- Iterative Improvement: The results show a clear progression in CT number accuracy and artifact reduction with increasing DIRA iterations, underscoring the benefit of iterative reconstruction for quantitative imaging.
- Material-Specific Accuracy: DIRA's ability to provide accurate CT numbers for diverse materials (metals, plastics, bone-like materials) is crucial for its application in various imaging scenarios.
Key Results
- Rod Phantom: By the sixth iteration, DIRA yielded CT numbers for the rod phantom that closely matched tabulated reference values. Specifically, for the aluminum rod, DIRA at iteration 6 showed a clear progression towards tabulated attenuation values, effectively correcting the downward curvature seen at iteration 0.
- Head Phantom: DIRA demonstrated strong performance on the anthropomorphic head phantom. The CT numbers for bone and teeth showed a significant increase with iterations, approaching specified values. Soft tissue homogeneity also improved, particularly in the tongue and neck regions.
- Artifact Suppression: DIRA successfully suppressed beam-hardening artifacts between dissimilar materials (e.g., aluminum and PTFE rods), an improvement over previous methods.
- Noise Levels: DIRA achieved noise levels comparable to conventional filtered backprojection (FBP) but did not match the noise reduction performance of advanced algorithms like ADMIRE.
Limitations and Future Work
- Noise Reduction: The authors acknowledge that DIRA, in its current form, does not match the noise reduction capabilities of algorithms like ADMIRE. Future work will focus on incorporating noise reduction techniques inspired by these advanced methods.
- Helical Acquisitions: The current study focused on axial scans. Future work will extend the evaluation to helical acquisitions, which are more common in clinical practice.
- Iodine-Containing Tasks: The study did not specifically evaluate DIRA's performance with iodine-containing contrast agents, which are frequently used in DECT. Extending evaluations to such tasks is a future direction.
- Specification Accuracy: For the CIRS head phantom, the authors note that the precision of the provided specification is unknown, making direct comparison of soft tissue values less reliable than for the fabricated rod phantom.
Summary updated