End-to-End Model-based Deep Learning for Dual-Energy Computed Tomography Material Decomposition

TL;DR

A deep learning procedure for quantitative material decomposition which directly convert the CT projection data into material images and shows the effectiveness of the proposed direct E2E-DEcomp method on the AAPM spectral CT dataset compared with state of the art supervised deep learning networks.

Abstract

Dual energy X-ray Computed Tomography (DECT) enables to automatically decompose materials in clinical images without the manual segmentation using the dependency of the X-ray linear attenuation with energy. In this work we propose a deep learning procedure called End-to-End Material Decomposition (E2E-DEcomp) for quantitative material decomposition which directly convert the CT projection data into material images. The algorithm is based on incorporating the knowledge of the spectral model DECT system into the deep learning training loss and combining a data-learned prior in the material image domain. Furthermore, the training does not require any energy-based images in the dataset but rather only sinogram and material images. We show the effectiveness of the proposed direct E2E-DEcomp method on the AAPM spectral CT dataset (Sidky and Pan, 2023) compared with state of the art supervised deep learning networks.

Figures (4)

• Figure 1: End-to-End Material Decomposition (E2E-DEcomp) material decomposition training workflow. The network is obtained by unfolding $K$ iterations of the algorithm and training end-to-end from energy sinograms $\mathbf{y}$ to material images $\mathbf{x}^K$. Each iteration is constituted by the data consistency block $\mathbb{DC}$ and the denoising module $\mathbb{D}_{\bm\rho}$ whose trainable parameters $\bm\rho$ are shared through the iterations.
• Figure 2: Qualitative comparison between the material decomposition for adipose using E2E-DEcomp and FBP using different number of angular projections.
• Figure 3: Comparison of DE-MoDL and FBP for 2 materials decomposition using noisy DECT acquisition with photon counts $I_0 = 10^5$. The PSNR metric is calculated for different number of DECT angular projections.
• Figure 4: Comparison of the PSNR training error between the FBP ConvNet and the E2E-DEcomp algorithms.