A foundation AI model enhances electron microscopy image analysis
Du, M.; Wang, Y.; Xie, L.; Deng, G.; Guo, J.; Han, B.; Chen, Z.-H.; Rui, C.; Han, J.; Chen, Y.; Zhao, Y.; Cao, R.; Wang, F.; Li, K.; Wang, Y.; He, Y.; feng, x.
Show abstract
Electron microscopy (EM) and advanced volume EM have extensive applications in deciphering cellular ultrastructures for life sciences. However, quality issues of EM images significantly impede the precise analysis and discovery of nanoscale biological structures. Here, we present an unsupervised Foundation Model for Five Tasks (DF5T) model for image enhancement through denoising, deblurring, super-resolution, two-dimensional (2D) inpainting, and three-dimensional (3D) isotropic restoration. DF5T was trained on an extensive dataset of multi-source membrane-bound organelle EM images, consisting of a total of over 2.25 million images. On unseen data, DF5T outperformed the existing state-of-the-art models in all five tasks. Furthermore, DF5T substantially compensates for missing three-dimensional structural information and reveals significant ultrastructural alterations in organelle geometry following chemical treatment. We demonstrate that DF5T significantly enhances electron microscopy images quality by improving the accuracy of downstream organelle segmentation and 3D structural restoration, thereby contributing to future advances in biological research.
Matching journals
The top 2 journals account for 50% of the predicted probability mass.
Similar papers in this journal
- Topaz-Denoise: general deep denoising models for cryoEM and cryoET 96%
- Beam image-shift accelerated data acquisition for near-atomic resolution single-particle cryo-electron tomography 96%
- Minutes-timescale 3D isotropic imaging of entire organs at subcellular resolution by content-aware compressed-sensing light-sheet microscopy 96%
Similar papers in this journal
- High-throughput, label-free and slide-free histological imaging by computational microscopy and unsupervised learning 95%
- Fast, 3D isotropic imaging of whole mouse brain using multi-1 angle-resolved subvoxel SPIM 95%
- Information-Distilled Generative Label-Free Morphological Profiling Encodes Cellular Heterogeneity 95%
Similar papers in this journal
Similar papers in this journal
- Resolution Enhancement with a Task-Assisted GAN to Guide Optical Nanoscopy Image Analysis and Acquisition 97%
- Construction of a 3D whole organism spatial atlas by joint modeling of multiple slices 93%
- Inferring spatial single-cell-level interactions through interpreting cell state and niche correlations learned by self-supervised graph transformer 92%
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.