SelfCryo: J-Invariant Volume Shuffle for Self-Supervised Cryo-Electron Tomogram Denoising on Single Noisy Volume
Liu, Xiwei
Liu, Xiwei
Author
Liu, Xiwei
Supervisor
Ho, Qirong
Department
Machine Learning
Embargo End Date
2025-05-30
Type
Thesis
Date
2025
License
Language
English
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
Cryo-Electron Tomography (Cryo-ET) enables detailed 3D visualization of cellular structures in nearnative states but suffers from low signaltonoise ratio due to imaging constraints. Traditional denoising methods and supervised learning approaches often struggle with complex noise patterns and the lack of paired training datasets. Self-supervised denoising methods, which utilize noisy input itself as a target, offer a promising solution; however, existing Cryo-ET selfsupervised denoising methods face significant challenges due to losing information during training and the learned incomplete noise patterns. To address these limitations, we propose SelfCryo, a novel selfsupervised learning model for Cryo-ET volumetric images denoising using a single noisy volume. SelfCryo features a Ushape J invariant blind spot network (BSN) with sparse centrally masked convolutions, dilated channel attention (DCA) blocks, and volumeunshuffle/shuffle technique. This architecture effectively expands receptive fields, enhances feature learning, and preserves fine structural details, significantly improving denoising performance. Extensive experiments on simulated and real world Cryo ET datasets demonstrate that SelfCryo outperforms existing selfsupervised Cryo-ET denoising methods, achieving superior noise suppression and structural preservation. Our findings highlight the potential of SelfCryo in advancing Cryo-ET data processing for structural biology research, providing a robust framework for improving tomographic reconstructions. The implementation is publicly available at =HYPERLINK("https://github.com/Xiwei-web/SelfCryoET", "https://github.com/Xiwei-web/SelfCryoET") CryoET.
Citation
Xiwei Liu, “SelfCryo: J-Invariant Volume Shuffle for Self-Supervised Cryo-Electron Tomogram Denoising on Single Noisy Volume,” Master of Science thesis, Machine Learning, MBZUAI, 2025.
Source
Conference
Keywords
Cryo-ET, Self-supervised, Denoising, Blind Spot Network, J-invariant