Room: Karl Dean Ballroom C
Purpose: In addition to providing electron densities, Dual energy CT (DECT) has the potential to differentiate material compositions. For Multi-Material Decomposition (MMD), due to the ill-defined problem, existing methods are limited in the number of materials that can be accurately separated and suffer from substantially amplified noise. In this study, we proposed a novel integrated MMD method exploiting the intrinsic sparsity on the number of materials that simultaneously present in the same voxel.
Methods: The proposed MMD was formulated as an optimization problem with a quadratic data fidelity term, along with an isotropic total variation term to regulate image smoothness, and a non-convex penalty function to promote sparsity in the decomposition domain. The mass and volume conservation rule was imposed as a constraint that the material component vector for each pixel is within the probability simplex. The optimization problem was solved using an accelerated primal-dual splitting approach with line search for non-convex problems. The proposed method with different penalty functions was compared with the classic direct inversion method on a digital phantom, Catphan 600 phantom, and one pelvis patient data.
Results: Compared with the direct inversion method, the proposed framework is able to distinctly separate the CT image into up to five basis materials plus air while controlling noise. The non-diagonal elements of the Normalized Cross Correlation (NCC) matrix are close to 0, showing little or no cross-talks between two different materials. The mean square error of the measured electron densities was reduced by 72.6%. Across all datasets, the proposed method improved the average Volume Fraction (VF) accuracy from 68.1% to 99.8%, and increased the diagonality of the NCC matrix from 0.80 to 0.95.
Conclusion: The novel MMD performs material decomposition into multiple basis materials with high accuracy and low cross-talks as compared with the classic direct inversion method.
Funding Support, Disclosures, and Conflict of Interest: This research is supported by NIH U19AI067769, DE-SC0017687, NIH R21CA228160, DE-SC0017057, NIH R44CA183390, NIH R43CA183390, and NIH R01CA188300.