An End-to-End Implicit Neural Representation Architecture for Medical Volume Data
2025

End-to-End Neural Network for Medical Volume Data Compression

Sample size: 750 publication 10 minutes Evidence: high

Author Information

Author(s): Sheibanifard Armin, Yu Hongchuan, Ruan Zongcai, Zhang Jian J.

Primary Institution: NCCA, Bournemouth University, Poole, United Kingdom

Hypothesis

Can an end-to-end architecture improve compression rates and reduce GPU memory usage for medical volume data?

Conclusion

The proposed architecture achieves a high compression rate of up to 97.5% while maintaining superior reconstruction accuracy.

Supporting Evidence

  • The architecture reduces GPU memory requirements and processing time.
  • Experimental results show a Peak Signal-to-Noise Ratio (PSNR) of 40.05 dB and Structural Similarity Index (SSIM) of 0.96.
  • The method is applicable to various medical imaging tasks.

Takeaway

This study created a smart system that makes big medical images smaller without losing important details, making it easier to store and use them.

Methodology

The study used a three-module architecture consisting of downsampling, implicit neural representation, and super-resolution to compress medical volume data.

Limitations

The architecture requires significant retraining for different datasets and the trade-off point method is time-consuming.

Participant Demographics

Patients diagnosed with glioblastoma or lower-grade glioma.

Digital Object Identifier (DOI)

10.1371/journal.pone.0314944

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication