Computational phase imaging for biomedical applications

Nguyen, Tan Huu

Permalink

https://hdl.handle.net/2142/95321 Copy

Description

Title

Computational phase imaging for biomedical applications

Author(s)

Nguyen, Tan Huu

Issue Date

2016-11-23

Director of Research (if dissertation) or Advisor (if thesis)

Popescu, Gabriel

Doctoral Committee Chair(s)

Popescu, Gabriel

Committee Member(s)

• Do, Minh N.
• Forsyth, David A.
• Kajdacsy-Balla, André A.

Department of Study

Electrical & Computer Eng

Discipline

Electrical & Computer Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

Ph.D.

Degree Level

Dissertation

Keyword(s)

• Quantitative Phase Imaging
• Holography
• Machine learning
• Imaging
• Optics
• Signal processing

Abstract

When a sample is illuminated by an imaging field, its fingerprints are left on the amplitude and the phase of the emerging wave. Capturing the information of the wavefront grants us a deeper understanding of the optical properties of the sample, and of the light-matter interaction. While the amplitude information has been intensively studied, the use of the phase information has been less common. Because all detectors are sensitive to intensity, not phase, wavefront measurements are significantly more challenging. Deploying optical interferometry to measure phase through phase-intensity conversion, quantitative phase imaging (QPI) has recently gained tremendous success in material and life sciences. The first topic of this dissertation describes our effort to develop a new QPI setup, named transmission Spatial Light Interference Microscopy (tSLIM), that uses the twisted nematic liquid-crystal (TNLC) modulators. Compared to the established SLIM technique, tSLIM is much less expensive to build than its predecessor (SLIM) while maintaining significant performance. The tSLIM system uses parallel aligned liquid-crystal (PANLC) modulators, has a slightly smaller signal-to-noise Ratio (SNR), and a more complicated model for the image formation. However, such complexity is well addressed by computing. Most importantly, tSLIM uses TNLC modulators that are popular in display LCDs. Therefore, the total cost of the system is significantly reduced. Alongside developing new imaging modalities, we also improved current QPI imaging systems. In practice, an incident field to the sample is rarely perfectly spatially coherent, i.e., plane wave. It is generally partially coherent; i.e., it comprises of many incoherent plane waves coming from multiple directions. This illumination yields artifacts in the phase measurement results, e.g., halo and phase-underestimation. One solution is using a very bright source, e.g., a laser, which can be spatially filtered very well. However, the laser comes at the expense of speckles, which degrades image quality. Therefore, solutions purely based on physical modeling and computations to remove these artifacts, using white-light illumination, are highly desirable. Here, using physical optics, we develop a theoretical model that accurately explains the effects of partial coherence on image information and phase information. The model is further combined with numerical processing to suppress the artifacts, and recover the correct phase information. The third topic is devoted to applying QPI to clinical applications. Traditionally, stained tissues are used in prostate cancer diagnosis instead. The reason is that tissue samples used in diagnosis are nearly transparent under bright field inspection if unstained. Contrast-enhanced microscopy techniques, e.g., phase contrast microscopy (PC) and differential interference contrast microscopy (DIC), can render visibility of the untagged samples with high throughput. However, since these methods are intensity-based, the contrast of acquired images varies significantly from one imaging facility to another, preventing them from being used in diagnosis. Inheriting the merits of PC, SLIM produces phase maps, which measure the refractive index of label-free samples. However, the maps measured by SLIM are not affected by variation in imaging conditions, e.g., illumination, magnification, etc., allowing consistent imaging results when using SLIM across different clinical institutions. Here, we combine SLIM images with machine learning for automatic diagnosis results for prostate cancer. We focus on two diagnosis problems of automatic Gleason grading and cancer vs. non-cancer diagnosis. Finally, we introduce a new imaging modality, named Gradient Light Interference Microscopy (GLIM), which is able to image through optically thick samples using low spatial coherence illumination. The key benefit of GLIM comes from a large numerical aperture of the condenser, which is 0.55 NA, about five times higher than that in SLIM. GLIM has an excellent depth sectioning when recording three-dimensional information of the susceptibility of the sample. We also introduce a model for the image formation of GLIM with an implication that a simple filtering step in the transverse dimension can dramatically improve the sectioning in the axial dimension. With GLIM, one can measure accurately the surface area, volume, and dry mass of a variety of biological samples, ranging from cells that are about tens of microns thick to bovine embryos that are hundreds of microns thick.

Graduation Semester

2016-12

Type of Resource

text

Permalink

http://hdl.handle.net/2142/95321

Copyright and License Information

Copyright 2016 Tan H. Nguyen

Owning Collections