Neural network for estimation of optical characteristics of optically active and turbid scattering media

TL;DR

This paper introduces a novel pipeline combining Monte Carlo simulation and deep neural networks to improve the accuracy of optical coherence tomography (OCT) imaging in turbid biological media, addressing inherent scattering challenges.

Contribution

The paper presents a new approach that integrates Monte Carlo simulation with deep learning to enhance OCT imaging quality without hardware modifications.

Findings

Improved estimation of optical properties in scattering media.

Enhanced OCT image accuracy through neural network correction.

Demonstrated effectiveness of the combined simulation and deep learning pipeline.

Abstract

One native source of quality deterioration in medical imaging, and especially in our case optical coherence tomography (OCT), is the turbid biological media in which photon does not take a predictable path and many scattering events would influence the effective path length and change the polarization of polarized light. This inherent problem would cause imaging errors even in the case of high resolution of interferometric methods. To address this problem and considering the inherent random nature of this problem, in the last decades some methods including Monte Carlo simulation for OCT was proposed. In this approach simulation would give us a one on one comparison of underlying physical structure and its OCT imaging counterpart. Although its goal was to give the practitioners a better understanding of underlying structure, it lacks in providing a comprehensive approach to increase the accuracy and imaging quality of OCT imaging and would only provide a set of examples on how imaging method might falter. To mitigate this problem and to demonstrate a new approach to improve the medical imaging without changing any hardware, we introduce a new pipeline consisting of Monte Carlo simulation followed by a deep neural network.