Combined iterative reconstruction and image-domain decomposition for dual energy CT using total-variation...

Area of Science:

• Medical Imaging
• Computed Tomography
• Image Processing

Background:

• Dual-energy CT (DECT) enables material decomposition and energy-selective imaging.
• Conventional DECT faces challenges with noise accumulation and signal cancellation during decomposition, degrading image quality.
• Existing noise reduction methods in DECT are often applied post-reconstruction, neglecting inherent statistical properties.

Purpose of the Study:

• To develop an iterative algorithm that integrates DECT image reconstruction and material decomposition.
• To minimize image noise and improve signal-to-noise ratio in DECT without compromising spatial resolution.
• To address the instability and noise accumulation issues inherent in standard DECT decomposition processes.

Main Methods:

• An iterative optimization framework was formulated, balancing data fidelity and total variation of decomposed images.
• The algorithm iteratively combines reconstruction and decomposition, promoting noise correlation in the reconstructed CT images.
• Performance was evaluated using phantom studies, comparing against conventional denoising and iterative reconstruction techniques.

Main Results:

• The proposed method achieved a one-order-of-magnitude reduction in noise standard deviation while preserving spatial resolution.
• High noise correlation in reconstructed images prevented noise accumulation during decomposition.
• Total variation regularization demonstrated superior edge preservation compared to other regularization methods, reducing electron density estimation error from 9.5% to 7.1%.

Conclusions:

• A practical, integrated iterative approach for DECT reconstruction and material decomposition has been presented.
• The method significantly enhances DECT imaging by improving decomposition accuracy, noise reduction, and spatial resolution.
• This integrated framework offers a superior alternative to existing DECT image processing techniques.