Resolution enhancement and noise suppression of optical coherence tomography via deep learning

Abstract

Optical coherence tomography (OCT) is an optical medical imaging modality that provides non-invasive three-dimensional imaging with high-speed and high-sensitivity in vivo. However, systemic limitations inherently arising from the basic operating principle of OCT degrade the resolution and SNR of images and limit their functionality as diagnostic tools. Recently, various hardware or software-based studies have been conducted to overcome these limitations, and one of them is application of deep learning that has been proven performance in diverse imaging fields. However, previous studies did not mainly utilize the raw signals and only demonstrated performance for specific systems and samples. Therefore, the performance improvement was limited and it was difficult to prove expandability to general-purpose applications. In this study, a deep learning-based processing for resolution enhancement and noise suppression was proposed based on theoretical understanding of optical coherence tomography systems and acquired signals, and performance is demonstrated for various samples. Through this, it is expected to overcome the limitations of existing optical consistency tomography and expand versatility as a medical diagnostic technology.