Deep learning-based denoising for improved dose efficiency in EDX tomography of nanoparticles

Area of Science:

• Materials Science
• Nanotechnology
• Analytical Chemistry

Background:

• Energy-dispersive X-ray spectroscopy (EDX) combined with electron tomography offers 3D elemental distribution analysis in nanomaterials.
• Current EDX tomography is limited by long acquisition times and high electron doses, impacting signal-to-noise ratio and material integrity.

Purpose of the Study:

• To develop an intelligent denoising method using deep learning to overcome EDX tomography limitations.
• To improve the trade-off between reconstruction quality, acquisition time, and radiation dose for EDX tomography.

Main Methods:

• Implementation of a deep learning methodology for intelligent denoising of experimental EDX data.
• Training the deep learning model with a focus on nanoparticle-like objects and noisy EDX signals.
• Quantitative analysis comparing the proposed method with classical denoising approaches.

Main Results:

• The deep learning denoising approach significantly enhances performance compared to classical methods.
• The method allows for improved 3D EDX reconstructions with reduced acquisition times and electron doses.
• Demonstrated effectiveness for electron beam-sensitive materials and nanoparticle transformation studies.

Conclusions:

• Deep learning-based denoising offers a powerful solution to enhance EDX tomography.
• This advancement enables more efficient and less damaging 3D elemental analysis of nanomaterials.
• Facilitates detailed investigation of sensitive materials and dynamic processes in nanoparticles.