Low-Dose Sparse-View HAADF-STEM-EDX Tomography of Nanocrystals Using Unsupervised Deep Learning

Area of Science:

• Materials Science
• Nanotechnology
• Electron Microscopy

Background:

• High-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) coupled with energy dispersive X-ray (EDX) spectroscopy provides elemental and structural information of nanoparticles.
• Deep learning methods show promise for 3D tomographic reconstruction, but typically require extensive high-quality training data, which is difficult to obtain due to electron beam-induced sample damage.
• Existing methods struggle with low-dose and sparse-view data, limiting tomographic reconstruction capabilities.

Purpose of the Study:

• To develop an unsupervised deep learning approach for HAADF-STEM-EDX tomography.
• To enable accurate 3D reconstruction of nanoparticles under low-dose and sparse-view conditions.
• To overcome the limitations of supervised learning and data acquisition challenges in electron tomography.

Main Methods:

• An unsupervised deep learning framework for HAADF-STEM-EDX tomography was developed.
• A HAADF-constrained unsupervised denoising method was proposed to enhance EDX image quality under low-dose conditions.
• An unsupervised view enrichment scheme in the projection domain was introduced to facilitate extreme sparse-view tomographic reconstruction.

Main Results:

• The proposed method successfully performed high-quality 3D tomographic reconstruction of quantum dots.
• Effective denoising of EDX data was achieved even under low-dose imaging.
• Accurate reconstruction was demonstrated using as few as three projection views.

Conclusions:

• The developed unsupervised deep learning method significantly advances HAADF-STEM-EDX tomography.
• This approach enables robust 3D nanoparticle imaging in challenging low-dose and sparse-view scenarios.
• The method holds potential for broader applications in materials characterization and nanotechnology where data acquisition is limited.