Compressive hyperspectral single-pixel imaging for lifetime imaging and tomography applications
Author
Ochoa, Marien I.Other Contributors
Intes, Xavier; Lesage, Frederic; Marcu, Laura; Pogue, Brian W.; Barroso, Margarida; Wang, Ge, 1957-; Yan, Pingkun;Date Issued
2022-08Subject
Biomedical engineeringDegree
PhD;Terms of Use
This electronic version is a licensed copy owned by Rensselaer Polytechnic Institute (RPI), Troy, NY. Copyright of original work retained by author.; Attribution-NonCommercial-NoDerivs 3.0 United StatesMetadata
Show full item recordAbstract
The present document summarizes the overall goal, specific objectives, preliminary results,current conclusions as well as the future aims to be addressed for completion of a PhD degree in Biomedical Engineering and is submitted to the Doctoral Committee to fulfill the PhD defense exam requisite. The main goal of this thesis project is to improve the quality and efficiency of single pixel based hyperspectral data acquisition and processing algorithms to further its utility for planar widefield hyperspectral lifetime imaging of extrinsic and intrinsic fluorescence, as well as tomographic reconstructions of absorption contrasts. For planar lifetime imaging, deep learning algorithms are developed to improve and facilitate both the intensity and lifetime imaging reconstruction processes. The reconstruction algorithms are optimized to extend data compression thereby shortening experimental acquisition time while maintaining image quality. Experimental validation has been accomplished in silico and with both extrinsic in vivo and intrinsic in vitro fluorescent markers. The imaging of intrinsic markers is a new avenue proposed to leverage the hyperspectral features that can be acquired with a single-pixel arrangement. Additionally, a deep learning algorithm has been developed to disentangle spectral overlaps of the marker’s emissions by using both intensity and lifetime information. This DL framework has been applied for the retrieval of relative abundance coefficients and further extended to deliver fluorophore concentrations. For widefield hyperspectral tomography, compressive sensing is used together with hyperspectral and time resolved data types for two different deep learning frameworks that aim at retrieving absorption contrast values and their spatial distribution, as well as concentration, without the need for an ill-posed inverse solved solution. The deep learning approaches, together with application driven optimization of the single-pixel optical setup, aim to display the multiple advantages of single-pixel hyperspectral strategies coupled to deep learning frameworks for planar and tomographic imaging.;Description
August 2022; School of EngineeringDepartment
Dept. of Biomedical Engineering;Publisher
Rensselaer Polytechnic Institute, Troy, NYRelationships
Rensselaer Theses and Dissertations Online Collection;Access
CC BY-NC-ND. Users may download and share copies with attribution in accordance with a Creative Commons Attribution-Noncommercial-No Derivative Works 3.0 license. No commercial use or derivatives are permitted without the explicit approval of the author.;Collections
Except where otherwise noted, this item's license is described as CC BY-NC-ND. Users may download and share copies with attribution in accordance with a Creative Commons
Attribution-Noncommercial-No Derivative Works 3.0 license. No commercial use or derivatives
are permitted without the explicit approval of the author.