Comparison of deep learning approaches for extreme low-SNR image restoration
GigaScience · 2026
DOI: 10.1093/gigascience/giag071 ↗Kurz gefasst
Vergleichsstudie in der Zeitschrift GigaScience, 2026. Verfasst von Nasreen Elizabeth Buhn und 5 weiteren Autor:innen. Beteiligte Einrichtung: Biological Sciences Department, California Polytechnic State University, 1 Grand Ave, San Luis Obispo, CA 93407-0401, USA. Hauptthemen laut MEDLINE: Deep Learning, Signal-To-Noise Ratio, Image Processing, Computer-Assisted. Die Arbeit wurde bisher 4-mal zitiert (Crossref) und nennt 38 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: NIGMS NIH HHS, NIH HHS, National Institute of General Medical Sciences.
Abstract (Originalsprache)
Background
Live-cell fluorescence microscopy enables the study of dynamic cellular processes. However, fluorescence microscopy can damage cells and disrupt these dynamic processes through photobleaching and phototoxicity. Reducing a sample’s light exposure mitigates the effects of photobleaching and phototoxicity but results in low signal-to-noise ratio (SNR) images. Deep learning provides a solution for restoring these low-SNR images. However, these deep learning methods require large, representative datasets for training, testing, and benchmarking, as well as substantial GPU memory, particularly for denoising large images.
Results
We present a new fluorescence microscopy dataset designed to expand the range of imaging conditions and specimens currently available for evaluating denoising methods. The dataset contains 17,568 paired high/low-SNR images across 15 sub-datasets that vary in specimen, imaging modality, objective, staining type, excitation wavelength, and exposure time. We evaluated five state-of-the-art deep learning denoising models on the dataset, including supervised, unsupervised, and zero-shot techniques. We also developed an image stitching method that enables large images to be processed in smaller crops and reconstructed.
Conclusions
Our dataset provides a diverse benchmark for evaluating deep learning denoising methods, and our stitching method provides a solution to GPU memory constraints encountered when processing large images. Among the evaluated deep learning models, the supervised Transformer-based model had the best denoising performance but required the longest training time.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Biological Sciences Department, California Polytechnic State University, 1 Grand Ave, San Luis Obispo, CA 93407-0401, USA.
- Department of Computer Science and Software Engineering, California Polytechnic State University, 1 Grand Ave, San Luis Obispo, CA 93407-0354, USA.
- UCCS BioFrontiers Center, University of Colorado at Colorado Springs, 1420 Austin Bluffs Parkway, Colorado Springs, CO 80918, USA.
ORCID-Profile: Nasreen Elizabeth Buhn, Guy M Hagen, Jonathan D Ventura
Förderung
- NIGMS NIH HHS
- NIH HHS
- National Institute of General Medical Sciences
- National Institutes of Health
Zitieren
APA
Buhn, N. E., Adunur, S. R., Hamilton, J., Levis, S., Hagen, G. M., & Ventura, J. D. (2026). Comparison of deep learning approaches for extreme low-SNR image restoration. GigaScience. https://doi.org/10.1093/gigascience/giag071
Vancouver
Buhn NE, Adunur SR, Hamilton J, Levis S, Hagen GM, Ventura JD. Comparison of deep learning approaches for extreme low-SNR image restoration. GigaScience. 2026. doi:10.1093/gigascience/giag071
BibTeX
@article{Buhn2026,
author = {Buhn, Nasreen Elizabeth and Adunur, Sriya Reddy and Hamilton, Joseph and Levis, Summer and Hagen, Guy M and Ventura, Jonathan D},
title = {Comparison of deep learning approaches for extreme low-SNR image restoration},
journal = {GigaScience},
year = {2026},
doi = {10.1093/gigascience/giag071}
}
EUVIMED