Cryo-electron microscope image denoising based on the geodesic distance.

Area of Science:

• Structural biology
• Biophysics
• Computational imaging

Background:

• Three-dimensional (3-D) reconstruction of viruses using electron cryomicroscopy (cryo-EM) requires accurate similarity assessment of 2-D projection image blocks.
• High-resolution cryo-EM projections are often obscured by significant noise, hindering precise analysis.
• Traditional methods like the Euler metric may not adequately capture the nuances of image block similarity in noisy cryo-EM data.

Purpose of the Study:

• To develop and validate a new method for measuring the similarity of image blocks in cryo-EM images.
• To address the challenge of noise in high-resolution cryo-EM projections.
• To enhance the accuracy of virus particle picking for improved 3-D reconstruction.

Main Methods:

• A novel approach utilizing the geodesic metric to measure the similarity between image blocks.
• Application of manifold learning techniques for image analysis.
• Testing the method on a dataset of 2243 cytoplasmic polyhedrosis virus (CPV) capsid particle images.

Main Results:

• The proposed geodesic metric method demonstrated superior performance compared to established techniques like Block-matching and 3-D filtering (BM3D), Stein's unbiased risk estimator (SURE), Bayes shrink, and K-means singular value decomposition (K-SVD).
• Achieved a peak signal-to-noise ratio (PSNR) of 45.65, indicating significant noise reduction.
• Successfully removed noise from cryo-EM images and improved the accuracy of particle picking.

Conclusions:

• The primary contribution is the application of geodesic distance for robust image block similarity measurement in cryo-EM.
• Manifold learning methods, specifically the proposed geodesic approach, are effective for denoising cryo-EM images.
• The enhanced image quality and particle picking accuracy facilitate more reliable virus 3-D reconstruction.