MATH-BASED APPROACHES FOR CONTRAST ENHANCEMENT OF BIOLOGICAL FEATURES BASED ON LABEL-FREE MULTI-MODAL MULTI-PHOTON BIO-IMAGES

Liu, Shitao

Permalink

https://hdl.handle.net/2142/124980 Copy

Description

Title

MATH-BASED APPROACHES FOR CONTRAST ENHANCEMENT OF BIOLOGICAL FEATURES BASED ON LABEL-FREE MULTI-MODAL MULTI-PHOTON BIO-IMAGES

Author(s)

Liu, Shitao

Issue Date

2021-12-01

Keyword(s)

Biomedical Optical Imaging; Unsupervised Learning;

Abstract

Label-free optical bioimaging techniques, such as Simultaneous Label-Free Autofluorescence Multi-harmonic Microscopy (SLAM), have been intensively developed for their advantages in enabling researchers to acquire molecular and structural information of living tissues promptly. However, the signalnoise ratio prevents us from precisely studying the areas of interest on the acquired images. Thus, an analytical tool is needed to help researchers further investigate the underlying biological patterns hidden behind the imaging channels. In this thesis, I pursue two approaches to study if latent features in labelfree multi-dimensional multi-modal bio-images can be enhanced using the raw image data. First, I study a mathematical approach to systematically study the linear and non-linear combinations of signal channels. I develop a quantitative model to describe the importance of channels and their abilities to provide extra information for further analysis. Secondly, I introduce techniques from generative machine learning area to bioimaging analysis. By applying generative models to the acquired image data, I establish an end-to-end pipeline to find the relationships between the variables, including the image channels as well as other latent variables, and the biological features of the areas of interest. On different scales, the two approaches both demonstrate their abilities to emphasize the latent features in the in vivo microenvironment. These results can be applied to provide biologists with extra information to help them more efficiently conduct analysis. I also propose several methods that can potentially improve the current results in future.

Type of Resource

text

Language

eng

Owning Collections