CryoEDNet: an edge-enhanced deep convolutional network for detail-preserving denoising of cryo-EM particle images
Year: 2026
Type: article
Abstract: Single-particle cryo-electron microscopy (cryo-EM) has revolutionized structural biology by enabling near-atomic reconstruction of biological macromolecules without crystallization. However, the low signal-to-noise ratio (SNR) of cryo-EM particle images limits the accuracy of downstream three-dimensional (3D) reconstruction. Existing denoising algorithms may blur structural boundaries, remove fine details, or provide limited benefits for 3D reconstruction. To address these problems, we propose CryoEDNet, an edge-enhanced deep convolutional neural network for cryo-EM particle image denoising. CryoEDNet includes a multi-scale feature extraction module, an edge enhancement module, and a detail preservation module. The network is trained using a multi-constraint loss that considers pixel-level reconstruction, perceptual similarity, structural similarity, and frequency-domain consistency. Experiments on synthetic projections with Gaussian white noise at SNR=0.1 show improvements in image quality, projection orientation estimation, and 3D reconstruction results. Tests on three experimental cryo-EM datasets (EMPIAR-10028, EMPIAR-10219, and EMPIAR-10330) show that CryoEDNet improves particle representation, 2D class averaging, and reconstruction-related outcomes under the evaluated conditions. These results support the transferability of CryoEDNet across the evaluated experimental datasets, while broader validation across more diverse cryo-EM specimens and imaging conditions remains necessary.
Source: Scientific Reports
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,690
Award: G393882934
OpenAlex ID: W7235022400
