Adaptive optimization framework for accurate multi-material decomposition in dual-energy CT.

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Processing

Background:

• Dual-energy CT (DECT) enables material decomposition for quantitative analysis.
• Conventional methods face challenges with noise and accuracy in material fraction estimation.

Purpose of the Study:

• Introduce an Adaptive Optimization-based Multi-Material Decomposition with Total Nuclear Variation (AO-MMD-TNV) for DECT.
• Enhance accuracy and noise robustness in material fraction estimation.

Main Methods:

• Developed AO-MMD-TNV algorithm combining Huber data term, L1 sparsity, and Total Nuclear Variation (TNV) regularization.
• Employed adaptive weighting to balance physical fidelity, sparsity, and boundary coherence.
• Evaluated using digital, tissue characterization, and human-shaped phantoms, comparing with conventional MMD.

Main Results:

• Achieved 100% volume fraction accuracy (VFA) and zero standard deviation (STD) in a digital phantom.
• Demonstrated improved VFA in most regions of interest (ROIs) for the tissue characterization phantom.
• Showcased stable performance and superior quantitative consistency and noise robustness in a human-shaped phantom across multiple ROIs.

Conclusions:

• AO-MMD-TNV framework significantly enhances quantitative reliability and reduces noise in DECT.
• The method preserves anatomical boundaries and shows potential for precise DECT-based material quantification.
• Further optimization for low-contrast materials is suggested for broader clinical application.