GPU Accelerated 3D Tomographic Reconstruction and Visualization From Noisy Electron Microscopy Tilt-Series

Area of Science:

• Structural Biology
• Microscopy and Imaging
• Computational Science

Background:

• Electron microscopy (EM) is crucial for high-resolution structural biology.
• Reconstructing 3D tomograms from noisy tilt-series is challenging.
• Current methods often require separate denoising and reconstruction steps.

Purpose of the Study:

• To present a novel, integrated framework for 3D tomographic reconstruction and visualization.
• To automate denoising within the reconstruction process, eliminating manual algorithm selection.
• To provide visualization of reconstruction errors.

Main Methods:

• A proximal jointly-optimized approach for iterative reconstruction and denoising.
• Leveraging GPU parallelism for accelerated processing.
• Incorporating volume rendering for immediate visualization.
• Open-source framework extensible with new algorithms.

Main Results:

• Successful denoising and reconstruction of 3D tomograms from noisy EM tilt-series.
• Demonstrated flexibility with various reconstruction algorithms and regularizers.
• Enabled real-time visualization of reconstruction errors.
• Outperformed state-of-the-art methods in evaluations.

Conclusions:

• The novel framework offers an efficient and automated solution for 3D EM tomogram reconstruction.
• Integrated denoising and reconstruction simplify the workflow and improve results.
• Error visualization aids in assessing and improving reconstruction quality.