Dynamic compressed sensing for real-time tomographic reconstruction.

Area of Science:

• Materials Science
• Physics
• Biotechnology

Background:

• Electron tomography (ET) advances offer higher resolution and quality at reduced electron doses.
• Compressed sensing (CS) reconstructs nanoscale 3D volumes from undersampled data by exploiting signal sparsity.
• Current CS methods in ET are computationally intensive, with reconstruction times ranging from hours to days.

Purpose of the Study:

• To develop a dynamic compressed sensing (CS) framework for real-time 3D volume reconstruction in electron tomography.
• To enable interactive analysis and high-throughput interpretation of nanoscale specimens during data acquisition.
• To accelerate the reconstruction process and improve the fidelity of 3D tomograms.

Main Methods:

• Implementation of a dynamic compressed sensing framework for iterative 3D reconstruction.
• Utilizing scanning transmission electron microscopy (STEM) for data acquisition.
• Real-time updating of the 3D specimen structure as new projections are collected.

Main Results:

• Dynamic CS framework accelerates convergence speed by approximately 3-fold compared to conventional CS.
• Reconstruction error is reduced by 27% for nanoscale specimens.
• Interpretable 3D tomogram structure is achieved within approximately 33% of data acquisition, with fine details visible by ~66% completion.
• Reconstruction parameters can be adjusted dynamically without restarting the entire process.

Conclusions:

• Dynamic CS provides a significant advancement for electron tomography, enabling real-time, high-fidelity 3D nanoscale imaging.
• The framework facilitates interactive analysis and accelerates research workflows in materials science and nanotechnology.
• This approach overcomes the computational bottlenecks of traditional CS, paving the way for more efficient high-throughput studies.