Machine learning denoising of high-resolution X-ray nanotomography data.

Area of Science:

• Materials Science
• Biomedical Imaging
• Nanotechnology

Background:

• X-ray nanotomography provides high-resolution 3D imaging crucial for quantitative analysis.
• Image noise in nanotomography hinders accurate segmentation and detailed investigation.
• Conventional filters often introduce blurring, compromising structural integrity.

Purpose of the Study:

• To evaluate a self-supervised machine learning (ML) denoising technique for X-ray nanotomography data.
• To compare the ML approach against traditional filtering methods like median and nonlocal means filters.
• To demonstrate the efficacy of ML in noise reduction without structural distortion.

Main Methods:

• Application of a self-supervised denoising ML algorithm to high-resolution X-ray nanotomography datasets.
• Comparative analysis with optimized median and nonlocal means filters.
• Assessment of noise reduction and preservation of structural features.

Main Results:

• The ML denoising technique effectively eliminated noise from nanotomography data.
• The ML approach significantly outperformed conventional filters in noise removal.
• Crucially, the ML method preserved relevant structural features, avoiding blurring.

Conclusions:

• Self-supervised ML denoising is a highly efficient and powerful tool for nanotomography.
• This technique enables robust quantitative analysis by providing clear, non-blurred tomograms.
• ML offers a superior alternative to conventional filtering for advancing scientific research using nanotomography.