Author: Dale Black, David Clymer, Huanjun Ding, Hamidreza Khodajou Chokami, Sabee Molloi, Christine Vy Nguyen, Tim Sananikone, Alireza Shojazadeh, Randy Wang 👨🔬
Affiliation: Department of Radiological Sciences, University of California, University of California, Department of Radiological Sciences, University of California, Irvine, Department of Radiological Sciences, University of California, Irvine 🌍
Purpose: We present a novel multi-material decomposition (MMD) algorithm for accurate quantification of material concentration in reconstructed dual-energy CT images. This method addresses limitations of existing techniques, including (1) numerical instability from poorly conditioned material triplet triangles and (2) lack of automated triplet selection for different anatomical regions. These issues reduce decomposition accuracy and increase computational complexity. Our method achieves over tenfold improvement in accuracy compared to prior approaches.
Methods: The unstable inversion of the linear attenuation coefficient (LAC) matrix for basis material volume fractions was replaced with a direct LAC-space area-ratio calculation, enhancing numerical stability. A centroid-based triplet selection criterion automatically identified optimal material triplets, eliminating dependency on triplet order. Principal component analysis (PCA) was applied to homogenize triangle geometry in LAC space, further reducing instability. The algorithm was validated using phantom data (Gammex 472) and patient pulmonary CT angiography images from a CdTe photon-counting CT scanner (Siemens Naeotom Alpha).
Results: The method showed high accuracy in iodine quantification (mg/mL), with a mean relative error of 1.86% (max: 2.67%), compared to 16.39% (max: 18.00%) in prior methods. Unlike existing techniques, which yielded unstable results when triplet order changed, our centroid-based selection ensured consistent decomposition. PCA-driven triangle homogenization eliminated highly acute triangles, improving material quantification. Across six patient scans, the proposed method consistently improved material separation and reduced dependency on predefined triangle ordering, while previous methods exhibited higher sensitivity to beam hardening artifacts and inconsistencies in soft tissue decomposition.
Conclusion: By integrating PCA for triangle homogenization, centroid-based triplet selection, and area-based volume fraction calculation, our MMD algorithm enhances accuracy, preserves stability, and improves efficiency, providing a robust solution for multi-material decomposition in dual-energy CT imaging.