Deep-learning-based image compression for microscopy images: An empirical study
2024

Deep Learning for Compressing Microscopy Images

Sample size: 500 publication Evidence: high

Author Information

Author(s): Zhou Yu, Sollmann Jan, Chen Jianxu

Primary Institution: Leibniz-Institut für Analytische Wissenschaften – ISAS – e.V.

Hypothesis

Can deep-learning-based image compression methods outperform classic methods in preserving image quality for microscopy images?

Conclusion

Deep-learning-based image compression techniques significantly outperform classic methods with minimal impact on downstream analysis tasks.

Supporting Evidence

  • Deep-learning-based compression methods achieved higher compression ratios than classic methods.
  • Minimal influence on downstream tasks was observed when using deep-learning-based compression.
  • Classic methods like JPEG-2000-LOSSY showed poor performance in terms of image quality.
  • Training models with compressed data improved prediction accuracy.
  • 3D compression results indicated a significant quality downgrade without proper training.

Takeaway

This study shows that using smart computer programs can make pictures smaller without losing important details, which helps scientists share and store their images better.

Methodology

The study compared classic and deep-learning-based image compression methods and analyzed their impact on downstream tasks using microscopy images.

Limitations

The study primarily used pre-trained models and did not fine-tune them on the microscopy dataset, which may limit performance.

Digital Object Identifier (DOI)

10.1017/S2633903X24000151

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication