Machine learning and quantum optics : an approach to improve OCT capabilities [article]

Krzysztof A. Maliszewski, University Of Canterbury
2022
In machine learning, a neural network is trained using datasets to provide a prediction of a pre-defined parameter. Due to its advanced image analysis capabilities, machine learning was quickly embraced by the optics and photonics community, providing solutions to decades-long problems or opening paths to achieving completely new quality or performance capabilities. Optical Coherence Tomography (OCT) is a light-based method able to provide highquality images of the inside of semi-transparent
more » ... ects, such as eyes. Originally proposed three decades ago, it is now in all ophthalmologists' offices, allowing eye disease diagnoses. OCT reaches beyond object visualisation, allowing functional imaging, e.g. quantifying blood flows or helping assess the stiffness of objects. Due to this incredible versatility, OCT is continuously researched to provide even better image quality and newer and broader functional capabilities. To be able to push OCT forward, scientists keep turning to other science disciplines. Using quantum optics concepts, quantum OCT was created, which provides two very soughtafter advancements: a twofold resolution increase and cancellation of resolution-degrading dispersion effects. In another method called quantum-mimic OCT, the quantum OCT signal is mimicked with a computer algorithm and a traditional OCT signal as an input, allowing to obtain the majority of quantum OCT features in a fast and robust way. Unfortunately, the advantages of these two attractive methods are buried in artefacts, unwanted elements in the image that completely obscure it. The thesis aims to extend the imaging and functional capabilities of OCT through a novel application of machine learning in quantum OCT and quantum-mimic OCT. Firstly, neural network models are built and trained to efficiently remove the artefacts in quantum OCT, therefore enabling access to quantum OCT's beneficial features. Secondly, neural network models are created for quantum-mimic OCT to infer the distribution of chromatic dispersion within the object [...]
doi:10.26021/13616 fatcat:wdc6eqfzcfcq3g22o37s2of5jy