Optimization of Fast Non-Local Means Noise Reduction Algorithm Parameter in Computed Tomographic Phantom Images Using...

Area of Science:

• Medical Imaging
• Image Processing
• Biomedical Engineering

Background:

• Noise in computed tomography (CT) images degrades diagnostic accuracy.
• Non-local means (NLM) is a common software technique for noise reduction in medical imaging.
• Developing advanced noise reduction methods is crucial for improving CT scan reliability.

Purpose of the Study:

• To propose and evaluate a novel noise reduction algorithm based on fast non-local means (FNLMs) for CT images.
• To assess the performance of FNLMs using a 3D-printed phantom mimicking human brain tissue density.
• To quantitatively analyze the effectiveness of FNLMs in improving CT image quality metrics.

Main Methods:

• Developed a noise reduction algorithm utilizing fast non-local means (FNLMs).
• Applied the FNLM algorithm to CT images of a custom-made 3D-printed phantom.
• Quantitatively evaluated image quality using contrast-to-noise ratio (CNR), coefficient of variation (COV), and normalized noise power spectrum (NNPS).
• Optimized smoothing factors for FNLMs across various noise intensities.
• Compared FNLM performance against noisy images, local filters, and total variation algorithms.

Main Results:

• Optimized smoothing factors for FNLMs were determined for different noise levels (e.g., 0.08 at 0.001 noise intensity).
• FNLMs demonstrated superior noise reduction performance compared to local filters and total variation methods.
• Optimized FNLMs significantly enhanced image quality: CNR improved 6.53-16.34 times, COV improved 6.55-18.28 times, and NNPS improved by 10⁻² mm² on average compared to noisy images.
• The 3D-printed phantom effectively simulated human brain tissue for evaluation.

Conclusions:

• The proposed FNLM algorithm effectively reduces noise in CT images.
• FNLMs show significant potential for enhancing CT image quality, particularly in conjunction with anthropomorphic phantoms.
• This noise reduction approach can improve diagnostic accuracy in medical imaging applications.