Iterative image-domain decomposition for dual-energy CT.

Area of Science:

• Medical Imaging
• Image Processing
• Computational Science

Background:

• Dual-energy CT (DECT) enables material decomposition but faces challenges with noise degrading image quality.
• Existing denoising methods applied before or after decomposition do not fully leverage noise characteristics, leading to suboptimal performance.

Purpose of the Study:

• To develop an iterative image-domain decomposition method for DECT that effectively suppresses noise while preserving image quality.
• To improve the clinical utility of DECT by addressing signal-to-noise ratio degradation during material decomposition.

Main Methods:

• An iterative, least-squares estimation with smoothness regularization was formulated, incorporating the full variance-covariance matrix of decomposed images.
• Edge detection was used to preserve boundary sharpness, assigning smaller weights to edge pixels in the regularization term.
• The algorithm iteratively performs decomposition and noise suppression, unlike methods applied solely before or after decomposition.

Main Results:

• The proposed method reduced noise standard deviation by over 98% on a phantom, maintaining spatial resolution (8 pairs/cm).
• It achieved better low-contrast detectability compared to an edge-preserving algorithm and reduced electron density measurement error by over 97%.
• On an anthropomorphic head phantom, noise was reduced by over 97% without blurring critical structures.

Conclusions:

• The developed iterative method effectively combines noise suppression and material decomposition in DECT.
• By utilizing full noise statistical properties and edge predetection, the algorithm achieves superior noise reduction, high spatial resolution, and enhanced low-contrast detectability.