| Author: | Li, Huanhao |
| Title: | Deciphering information from random speckles via correlation and deep learning as well as its applications in biomedicine |
| Advisors: | Lai, Puxiang (BME) |
| Degree: | Ph.D. |
| Year: | 2021 |
| Subject: | Speckle Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Biomedical Engineering |
| Pages: | 122 pages : color illustrations |
| Language: | English |
| Abstract: | Multiple scattering dominates the propagation of light in complex media, such as biological tissues, and limits our vision within shallow layers or superficial depths. Under coherent optical illumination, seemingly random speckle patterns, rather than clear or blurred target images, are generated, where the optical information of the medium and the target are completely delocalized and scrambled. Extracting information of interest from speckles has significant scientific values and hence has been desired long yet considered challenging due to the randomized nature of speckles. My research in the past four years has mainly focused on new understandings of speckles and decipherment of speckle information via correlation and deep learning, aiming to extend human vision at depths in complex media. Accordingly, the thesis is organized through three major studies as follows. In the first study, explicitly correlated specklegram (within the memory effect) is characterized by the correlation coefficient among speckle patterns captured under different conditions to delineate the strength of photoacoustic (PA) signals generated within a scattering medium via an interferometry-free setting. This speckle contrast based approach provides a sensitivity comparable to those measured by traditional piezoelectric ultrasonic transducers. Such a plain non-contact specklegram-based method may find special interests for photoacoustic imaging when no direct contact is allowed or favored between the sample and the photoacoustic detector. In the second study, implicit speckle correlation is examined via deep learning since more complicated information is encoded within the speckle pattern itself. Special attentions are paid to recover the significantly defected speckles due to poor sampling, which cannot be processed by existing methods due to the lack of sufficient physical understanding or mathematical description. Experimentally, this learning-based platform allows one to recover the morphology and encoded information from 1) speckles sampled below ~14 times of the Nyquist criterion and 2) binarized speckles where only ~0.4% grains and information remained in the speckle field of view. More importantly, even though the training process for the speckle recovery is independent on any prior knowledge of the encoded information, the 'lost' information of the defected speckles can be visualized after the learning-based recovery. Such capabilities empower the learning-based framework a robust and promising mechanism to compress speckled signals at an ultra-low sampling setting, which can effectively conserve operating memory and data storage by a factor of 10^3, and hence greatly accelerate the speed of data acquisition and signal processing of massive speckled signals. In the third study, the scattering process is resembled via deep learning to explore how speckles encode information through speckle features, including location, profile, and brightness accuracy of the grains. As a preliminary study, it is inferred that, as the baseline for information encoding, the grain location is the primary feature. By adding the other two features, the shape/profile and the relative brightness of the speckle grains can be further differentiated. Consequently, the information capacity of speckles is considerably expanded, allowing more accurate and reliable information recovery from speckles. Whilst with a lot to be further explored, this pilot study opens a new avenue to reveal how information is delocalized and stored in complicated speckle patterns and how speckles can be decomposed effectively for analysis. In summary, this thesis explores to decipher speckles from perspectives of information extraction, interpolation, and morphological representation for encoded object information. These explorations may suggest additional understanding about speckles, which is complementary to conventional speckle statistics wisdom. With further development, it may provide a promising paradigm for efficient processing or deciphering of massive scattered optical signals, allowing one to see big and see clearly simultaneously within or through biological tissues. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/12378