Skip to content
DraconiticResearch
Search Sources API
203,234 studies31 sourcesUpdated 58 min ago
Peer-reviewedCC BYNature communications

ReScale4DL: balancing pixel and contextual information for enhanced bioimage segmentation.

  • Mariana G Ferreira1
  • Bruno M Saraiva1
  • António D Brito1,2
  • Mariana G Pinho2
  • Ricardo Henriques3,4
  • Estibaliz Gómez-de-Mariscal5,6,7
7 affiliations
  1. 1AI-driven Optical Biology Laboratory, Instituto de Tecnologia Química e Biológica António Xavier, Universidade Nova de Lisboa, Oeiras, Portugal.
  2. 2Bacterial Cell Biology Laboratory, Instituto de Tecnologia Química e Biológica António Xavier, Universidade Nova de Lisboa, Oeiras, Portugal.
  3. 3AI-driven Optical Biology Laboratory, Instituto de Tecnologia Química e Biológica António Xavier, Universidade Nova de Lisboa, Oeiras, Portugal. [email protected].
  4. 4UCL Laboratory for Molecular Cell Biology, University College London, London, UK. [email protected].
  5. 5AI-driven Optical Biology Laboratory, Instituto de Tecnologia Química e Biológica António Xavier, Universidade Nova de Lisboa, Oeiras, Portugal. [email protected].
  6. 6NOVA Institute for Medical Systems Biology (NIMSB), Universidade Nova de Lisboa, Lisboa, Portugal. [email protected].
  7. 7iNOVA4Health, Programme in Translational Medicine, Lisboa, Portugal. [email protected].
Read full textPDFDOICite

Abstract

Deep learning is the state-of-the-art approach for bioimage segmentation. However, it presents a paradox regarding image resolution: counterintuitively, deep learning segmentation performance can improve with lower image resolutions. This phenomenon is particularly significant in microscopy, where high-resolution acquisitions come with substantial costs in throughput, storage requirements and potential photodamage. We systematically evaluate how image resolution impacts segmentation by training popular architectures on datasets downsampled to 6-50% of their original resolution, mimicking lower-magnification acquisitions. Compared with models trained on native-resolution images, segmentation accuracy either improves (by up to 25% of mean Intersection over Union (IoU)) or degrades minimally (< 5% of mean IoU) when using images downsampled by up to fourfold (25% of the original resolution). Downsampling proportionally increases information throughput while reducing storage requirements and inference time. These findings provide practical guidelines for creating efficient, sustainable and cost-effective bioimaging pipelines that reduce data and computing needs while optimising microscopy techniques.

Related

  1. Metabolomics for Athlete Monitoring: An Evidence-Readiness Framework for Contextual Validation
    Donghai Lin, Yifen Chen, Caihua Huang +1
    MetabolitesPeer-reviewedCC BYFull textProteomics & metabolomicsEurope PMC
  2. Highlight: Balancing selection, more frequent than anticipated?
    Molecular biology and evolutionPeer-reviewedCC BY-NCFull textGeneral biochemistryEurope PMC
  3. ATHENA: a dynamic metabolic hub balancing carbon economy and energy supply in microbial synthesis
    Likun Guo, Xinjun Feng, Xingyu Zhang +6
    Nature CommunicationsPeer-reviewedCC BY-NC-ND 4.0Full textMolecular biology & gene regulationPMC
  4. PP2A targeting: balancing anticancer potential and neurological risk.
    Margarita Chigriai, Nina Kraskovskaya, Anastasia Bolshakova +3
    Frontiers in cell and developmental biologyPeer-reviewedCC BYFull textMolecular biology & gene regulationEurope PMCPMC
  5. MambaGLA-NPC: a 3D framework integrating CNN and state space models for robust multi-center nasopharyngeal carcinoma segmentation
    Wei Jin, Xiaobo Chen, Weifeng Jin +2
    Frontiers in genome editingPeer-reviewedCC BYFull textMolecular biology & gene regulationEurope PMC

Cite

BibTeXRIS
BibTeX
@article{ferreira2026rescale4dl,
  title = {{ReScale4DL: balancing pixel and contextual information for enhanced bioimage segmentation.}},
  author = {Mariana G Ferreira and Bruno M Saraiva and António D Brito and Mariana G Pinho and Ricardo Henriques and Estibaliz Gómez-de-Mariscal},
  journal = {Nature communications},
  year = {2026},
  month = sep,
  doi = {10.1038/s41467-026-77930-1},
  url = {https://europepmc.org/article/MED/42897196},
  note = {CC BY},
}
RIS
TY  - JOUR
TI  - ReScale4DL: balancing pixel and contextual information for enhanced bioimage segmentation.
AU  - Mariana G Ferreira
AU  - Bruno M Saraiva
AU  - António D Brito
AU  - Mariana G Pinho
AU  - Ricardo Henriques
AU  - Estibaliz Gómez-de-Mariscal
PY  - 2026
DA  - 2026/09/18
JO  - Nature communications
DO  - 10.1038/s41467-026-77930-1
UR  - https://europepmc.org/article/MED/42897196
AB  - Deep learning is the state-of-the-art approach for bioimage segmentation. However, it presents a paradox regarding image resolution: counterintuitively, deep learning segmentation performance can improve with lower image resolutions. This phenomenon is particularly significant in microscopy, where high-resolution acquisitions come with substantial costs in throughput, storage requirements and potential photodamage. We systematically evaluate how image resolution impacts segmentation by training popular architectures on datasets downsampled to 6-50% of their original resolution, mimicking lower-magnification acquisitions. Compared with models trained on native-resolution images, segmentation accuracy either improves (by up to 25% of mean Intersection over Union (IoU)) or degrades minimally (< 5% of mean IoU) when using images downsampled by up to fourfold (25% of the original resolution). Downsampling proportionally increases information throughput while reducing storage requirements and inference time. These findings provide practical guidelines for creating efficient, sustainable and cost-effective bioimaging pipelines that reduce data and computing needs while optimising microscopy techniques.
SN  - 2041-1723
ER  -