| About this Abstract |
| Meeting |
TMS Specialty Congress 2025
|
| Symposium
|
The 7th International Congress on 3D Materials Science (3DMS 2025)
|
| Presentation Title |
Optimizing CT Image Quality With Deep Learning for Enhanced 3D Materials Science Applications |
| Author(s) |
Parisa Asadi, Adrian Sarapata, Andriy Andreyev |
| On-Site Speaker (Planned) |
Parisa Asadi |
| Abstract Scope |
X-ray computed tomography (CT) is fundamental to non-destructive 3D imaging, though image quality is often hampered by noise. This study examines the benefits of FAST Mode image acquisition (ability to scan in 20 seconds) for rapid sample inspection and DeepRecon Pro 3D-image reconstruction, a deep learning-based noise reduction method adaptable to various imaging conditions. Using a U-Net architecture, DeepRecon Pro processes projection images or volumes to produce noise-suppressed, high-fidelity reconstructions, outperforming traditional techniques like FDK and non-local means (NLM). The latest DeepRecon version integrates synthetic priors and a two-stage training process, introducing noise-matching models that further enhance accuracy. Quantitative metrics, including mean square error (MSE) and structural similarity index (SSIM), highlight DeepRecon’s effectiveness across diverse noise intensities and projection counts. These advancements support faster, more accurate CT imaging for critical 3D materials science applications where precision and efficiency are paramount. |
| Proceedings Inclusion? |
Undecided |