DECT-MULTRA: Dual-Energy CT Image Decomposition With Learned Mixed Material Models and Efficient Clustering.

Area of Science:

• Medical Imaging
• Image Processing
• Computational Imaging

Background:

• Dual-energy computed tomography (DECT) offers material decomposition capabilities crucial for advanced imaging.
• Current image-domain decomposition methods using linear matrix inversion are susceptible to noise and artifacts, degrading material image quality.

Purpose of the Study:

• To propose DECT-MULTRA, an innovative image-domain DECT material decomposition method.
• To enhance material decomposition accuracy and image quality in DECT by mitigating noise and artifacts.

Main Methods:

• The DECT-MULTRA method integrates penalized weighted-least squares (PWLS) estimation with a mixed union of learned transforms (MULTRA) model.
• It employs pre-learned common-material and cross-material sparsifying transforms to capture material properties and dependencies.
• Optimization is achieved through an efficient alternating process involving image updates and sparse coding/clustering steps with closed-form solutions.

Main Results:

• DECT-MULTRA demonstrated superior material image quality compared to existing methods.
• The method achieved higher decomposition accuracy in both simulated (XCAT phantom) and real clinical (head data) datasets.
• Validation confirmed the effectiveness of the learned transforms and the optimization strategy.

Conclusions:

• DECT-MULTRA represents a significant advancement in image-domain DECT material decomposition.
• The proposed method effectively reduces noise and artifacts, leading to improved diagnostic information from DECT scans.
• This technique holds promise for enhancing various clinical applications of DECT imaging.