Joint Denoising of Cryo-EM Projection Images using Polar Transformers

Joakim Andén, Justus Sagemüller

Keywords: denoising, microscopy, tomography, attention, symmetries

TL;DR: Transformers can bring together information of multiple random-rotated images for better denoising.

Abstract:

Deep neural networks (DNNs) have proven powerful for denoising individual images, but there is a limit to the noise level they can handle.
In applications like cryogenic electron microscopy (cryo-EM), the noise level is extremely high but datasets contain hundreds of thousands of projections of the same molecule, each taken a different viewing direction.
This redundancy of information is useful in traditional denoising techniques known as class averaging methods, where images are clustered, aligned, and then averaged to reduce the noise level.
We present a neural network architecture based on polar representation of images and transformers that simultaneously clusters, aligns, and denoises cryo-EM projection images.
Results on synthetic data show accurate denoising performance using this architecture, with a relative mean squared error of $0.06$ at signal-to-noise (SNR) level of $0.05$, outperforming traditional filter-based methods by a factor of $2\times$.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 11936