Neural network optimization of Speckle-based phase-contrast X-ray imaging for mammography applications
X-ray Phase-Contrast Imaging (PCI) has revolutionized imaging in biomedical and materials science by enhancing contrast for soft tissues and subtle structural details that are challenging to detect with conventional absorption-based methods. Speckle-Based Imaging (SBI), a subset of PCI, leverages sp...
- Autores:
-
Naranjo Barros, Sofía
- Tipo de recurso:
- Trabajo de grado de pregrado
- Fecha de publicación:
- 2025
- Institución:
- Universidad de los Andes
- Repositorio:
- Séneca: repositorio Uniandes
- Idioma:
- eng
- OAI Identifier:
- oai:repositorio.uniandes.edu.co:1992/75726
- Acceso en línea:
- https://hdl.handle.net/1992/75726
- Palabra clave:
- Speckle-based imaging
Mammography
X-ray phase contrast
Single-exposure
Phase retrieval
Convolutional neural networks
Dose efficiency
Física
- Rights
- openAccess
- License
- Attribution 4.0 International
| Summary: | X-ray Phase-Contrast Imaging (PCI) has revolutionized imaging in biomedical and materials science by enhancing contrast for soft tissues and subtle structural details that are challenging to detect with conventional absorption-based methods. Speckle-Based Imaging (SBI), a subset of PCI, leverages speckle patterns created by random diffusers to retrieve phase information, offering improved soft tissue visualization without relying on highly coherent X-ray sources. This makes SBI particularly promising for mammography, where early detection of breast cancer is often limited by the low contrast between healthy and pathological tissues in traditional imaging techniques. Despite its advantages, SBI's reliance on multiple exposures for phase retrieval can increase radiation dose, posing a safety concern. This study seeks to optimize SBI for mammographic applications by integrating convolutional neural networks (CNNs) to enable single-exposure phase retrieval, significantly reducing radiation dose while maintaining high image quality. By adapting the CNN-based StrainNet-f model, the research explores the feasibility of applying CNNs in SBI for mammography. The optimized technique was validated computationally and experimentally by comparing retrieved images with theoretical benchmarks. The results provide a proof of principle for the application of CNNs in SBI, laying the groundwork for further research into developing safer and more efficient imaging techniques for breast cancer diagnostics. |
|---|