Physics-informed neural networks for optical fibre-based imaging
Publication Date
July 29, 2025
Creator
Abstract
Current medical imaging techniques show limitations in accurate early disease detection due to poor resolution, low contrast, and limited sensitivity to functional changes. Although optical imaging provides higher resolution, it cannot reach deep internal areas such as blood vessels, the brain, or organs like the pancreas. Ultra-thin optical fibre-based imaging systems offer a promising solution for high-resolution imaging in these hard-to-reach areas. However, severe optical distortions and noise in these systems degrade image quality, which makes clinical diagnosis more challenging.
This thesis aims to develop advanced optical fibre-based imaging technologies by utilising physics-informed neural networks (NNs) to reconstruct high-quality images from low-quality raw inputs. Traditional physical models, which are generally accurate, struggle with scalability and real-time performance. Neural Networks (NNs) which applied Artificial Intelligence (AI) techniques are difficult to handle the complex, large-scale data generated by fibre systems. I therefore proposed Physics-informed NNs to overcome these issues by embedding physical information into the model architecture, improving the generalization, interpretability, and accuracy while preventing the model from overfitting. This thesis presents two main contributions: (1) The application of physics-informed NNs to improve next-generation endoscopic imaging, and (2) Integrating biological information to use physics and biology-informed NNs applied to biological experiments using advanced microscopy for better image reconstruction.
The results proposed in this thesis demonstrate that physics-informed NNs are capable of enhancing image quality, showing great promises in improving early disease detection by reducing noise and enhancing contrast. This research high- lights the potential of physics-informed neural networks to advance optical fibre- based imaging technologies, paving the way for more effective and less invasive diagnostic tools. Ultimately, this work contributes to the development of next- generation medical imaging systems designed to improve early and accurate diagnosis of diseases.
Item Type
ethesis
Thesis Type
PhD
Supervisors
Subjects (LC)
Associated Schools / Departments
Department of Electrical and Electronic Engineering (UK)
eprints ID
81043
UoN Repository URI
Except where otherwise noted, this item's license is described as
File(s)![Thumbnail Image]()
Name
Thesis_corrected_amy.pdf
Type
Full-text
Description
Examined
Size
121.09 MB
Format
Adobe PDF
Checksum (MD5)
7c1bf0f108b5d6104aa10ded15b3c4ab