A Two-Step Framework for Multi-Material Decomposition of Dual Energy Computed Tomography from Projection Domain

Background and Purpose: Dual-energy computed tomography (DECT) utilizes separate X-ray energy spectra to improve multi-material decomposition (MMD) for various diagnostic applications. However accurate decomposing more than two types of material remains challenging using conventional methods. Deep learning (DL) methods have shown promise to improve the MMD performance, but typical approaches of conducing DL-MMD in the image domain fail to fully utilize projection information or are under computationally inefficient iterative setup. In this work, we present a clinical-applicable MMD (> 2) framework - rFast-MMDNet, operating with raw projection data in non-recursive setup, for breast tissue differentiation.Methods: rFast-MMDNet is a two-stage algorithm, including stage-one SinoNet to perform dual energy projection decomposition on tissue sinograms and stage-two FBP-DenoiseNet to perform domain adaptation and image post-processing. rFast-MMDNet was tested on a 2022 DL-Spectral-Challenge dataset, which includes 1000 pairs of training, 10 pairs of validation, and 100 pairs of testing images simulating dual energy fast kVp-switching fan beam CT projections of breast phantoms. MMD for breast fibroglandular, adipose tissues and calcification was performed. The two stages of rFast-MMDNet were evaluated separately and then compared with four noniterative reference methods including a direct inversion method (AA-MMD), an image domain DL method (ID-UNet), AA-MMD/ID-UNet + DenoiseNet and a sinogram domain DL method (Triple-CBCT).Results: Our results show that models trained from information stored in DE transmission domain can yield high-fidelity decomposition with averaged RMSE, MAE, negative PSNR, and SSIM of 0.004 ± 0, 0.001 ± 0, −45.027 ± 0.542, and 0.002±0 benchmarking to the ground truth, respectively. The inference time of rFast-MMDNet is < +1s. All DL methods generally led to more accurate MMD than AA-MMD. rFast-MMDNet outperformed Triple-CBCT, but both are superior to the image-domain based methods.Conclusions: A fast, robust, intuitive, and interpretable work-flow is presented to facilitate an efficient and precise MMD with input from projection domain.