Improving Phase-Contrast CT with Deep Learning Denoising
Author Information
Author(s): Wirtensohn Sami, Schmid Clemens, Berthe Daniel, John Dominik, Heck Lisa, Taphorn Kirsten, Flenner Silja, Herzen Julia
Primary Institution: Technical University of Munich
Hypothesis
Can self-supervised deep learning improve the denoising of grating-based phase-contrast computed tomography (gbPC-CT)?
Conclusion
The self-supervised deep learning network Noise2Inverse significantly enhances image quality in gbPC-CT, allowing for higher resolution while maintaining lower radiation doses.
Supporting Evidence
- Noise2Inverse can be trained on a single noisy tomogram, making it suitable for medical imaging applications.
- The application of N2I allows for higher resolution imaging while maintaining necessary contrast-to-noise ratios.
- N2I outperformed traditional denoising methods in improving image quality metrics.
- Using N2I can shift the break-even point in favor of gbPC-CT compared to conventional absorption-based CT.
Takeaway
This study shows that a smart computer program can help make better pictures of the inside of our bodies using less radiation.
Methodology
The study used a self-supervised deep learning network called Noise2Inverse to denoise images from gbPC-CT and compared its performance with other denoising methods.
Limitations
The study primarily focused on a single phantom and may not generalize to all clinical scenarios.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website