Unsupervised Cryo-EM Images Denoising and Clustering Based on Deep Convolutional Autoencoder and K-Means++
Published in IEEE Transactions on Medical Imaging, 2023
Abstract
Single-particle cryo-EM datasets contain large numbers of low signal-to-noise particle images, making denoising and unsupervised grouping important for downstream structure determination. This work develops an unsupervised framework based on a deep convolutional autoencoder and K-Means++ clustering to learn compact image representations, suppress noise, and separate particle images into meaningful groups without relying on manually labeled training data. The method targets practical cryo-EM preprocessing by combining learned feature extraction with balanced clustering, supporting image denoising and particle organization before later reconstruction and analysis stages.
Recommended citation: Dongxu Zhang, Yang Yan, Yulin Huang, Bowen Liu, Qingbing Zheng, Jun Zhang, Ningshao Xia. (2023). "Unsupervised Cryo-EM Images Denoising and Clustering Based on Deep Convolutional Autoencoder and K-Means++." IEEE Transactions on Medical Imaging, 42(5), 1509-1521. DOI: 10.1109/TMI.2022.3231626.
Publisher page