An effective noise reduction method for multi-energy CT images that exploit spatio-spectral features

Area of Science:

• Medical Imaging
• Image Processing
• Computed Tomography

Background:

• Multi-energy CT (MECT) offers advanced material differentiation capabilities.
• Image noise is a significant challenge in MECT, potentially hindering diagnostic accuracy.
• Developing effective noise reduction methods is crucial for optimizing MECT performance.

Purpose of the Study:

• To develop and evaluate an image-domain noise reduction method for MECT data.
• To assess the impact of the proposed method on various image quality metrics.
• To determine the clinical feasibility of the noise reduction technique.

Main Methods:

• A novel Multi-Energy Non-Local Means (MENLM) filter was developed, utilizing spatio-spectral features for robust noise reduction.
• The MENLM filter was tested on a photon counting CT system using phantom, swine, and cadaver datasets.
• Quantitative evaluations included noise level, noise power spectrum (NPS), spatial resolution (MTF, SSP), CT number accuracy, and material decomposition.

Main Results:

• MENLM achieved approximately 80% noise reduction while preserving NPS characteristics.
• High-contrast spatial resolution and CT number accuracy remained unaffected by the filtering.
• Material decomposition performance improved, and low-contrast detectability was enhanced in phantom studies.
• Clinical images showed improved visualization of vascular structures and gray/white matter differentiation.

Conclusions:

• MENLM effectively reduces noise in MECT images, enhancing material decomposition and detection of subtle features.
• The method maintains spatial and energy resolution, crucial for diagnostic accuracy.
• MENLM holds potential for reducing radiation dose and contrast media in MECT applications.