國立虎尾科技大學 |

Deep Learning in Label-Free Cell Sorting and Numerical Simulation With Reduced Training Data.

Record Type:	Language materials, manuscript : Monograph/item
Title/Author:	Deep Learning in Label-Free Cell Sorting and Numerical Simulation With Reduced Training Data./
Author:	Wang, Shen.
Description:	1 online resource (154 pages)
Notes:	Source: Dissertations Abstracts International, Volume: 85-07, Section: B.
Contained By:	Dissertations Abstracts International85-07B.
Subject:	Engineering. -
Online resource:	click for full text (PQDT)
ISBN:	9798381376784

Deep Learning in Label-Free Cell Sorting and Numerical Simulation With Reduced Training Data.
Wang, Shen.

Deep Learning in Label-Free Cell Sorting and Numerical Simulation With Reduced Training Data. - 1 online resource (154 pages)

Source: Dissertations Abstracts International, Volume: 85-07, Section: B.

Thesis (Ph.D.)--Lehigh University, 2024.

Includes bibliographical references

Many deep-learning-based problems depend on a significant amount and high quality of the dataset, while in the "small data" regime, less work and related applications have been established. The cost of microfluidic experiments and related fluid dynamics simulations limits the scale, predictability, and use in physics-conforming control systems for mobile and low-powered devices. On one hand, the feature extraction from data including images and sequential information in microfluidics is hard due to the potential noisy and limited scale. On the other hand, deep-learning-based simulations mostly rely on labeled training data. This dissertation presents innovative techniques based on data-driven approaches for better solutions in various experimental and computational applications. The first part of this dissertation presents an innovative technique for the detection of rare cells using a deep learning model on microscopic cell images. This proof-of-concept application demonstrates the potential for characterizing rare cells based on image data. Building on this success, we extend the application to an image-based cell sorting system that can identify distinct subpopulations of hematopoietic stem cells and multipotent progenitors. We also delve into the sensitivity of the deep learning-based classifier to the scale of the training dataset. In parallel, we tackle the problem of distortion correction in laser-based galvanometer fabrication systems. Leveraging a hybrid methodology that integrates computer vision algorithms and machine learning, we offer a cloud-based service that enables calibration and correction of lithography systems based on a set of calibration photos. Lastly, we establish a physics-informed model using a stacked U-Net architecture trained in a weakly-supervised manner on computationally-efficient, low-dimensional simulations. This model is created to generate numerical solutions for Navier-Stokes equations without the need for computationally expensive simulation results. An additional physics-informed model is developed to assist in the design of deterministic lateral displacement devices, indicating the potential for leveraging deep learning in microfluidic device design optimization. In essence, this dissertation bridges the gap between deep learning and microfluidics, bringing forth innovative solutions for cell characterization, system calibration, and computational problems within the "small data" regime.

Electronic reproduction.
Ann Arbor, Mich. :
ProQuest,
2024

Mode of access: World Wide Web

ISBN: 9798381376784Subjects--Topical Terms:

561152
Engineering.
Subjects--Index Terms:

Small dataIndex Terms--Genre/Form:

554714
Electronic books.

Deep Learning in Label-Free Cell Sorting and Numerical Simulation With Reduced Training Data.
LDR:03840ntm a22003977 4500 001 1148716
005 20240930100125.5
006 m o d
007 cr bn ---uuuuu
008 250605s2024 xx obm 000 0 eng d
020 $a 9798381376784
035 $a (MiAaPQ)AAI30635459
035 $a AAI30635459
040 $a MiAaPQ $b eng $c MiAaPQ $d NTU
100 1 $a Wang, Shen. $3 1107249
245 1 0 $a Deep Learning in Label-Free Cell Sorting and Numerical Simulation With Reduced Training Data.
264 0 $c 2024
300 $a 1 online resource (154 pages)
336 $a text $b txt $2 rdacontent
337 $a computer $b c $2 rdamedia
338 $a online resource $b cr $2 rdacarrier
500 $a Source: Dissertations Abstracts International, Volume: 85-07, Section: B.
500 $a Advisor: Liu, Yaling.
502 $a Thesis (Ph.D.)--Lehigh University, 2024.
504 $a Includes bibliographical references
520 $a Many deep-learning-based problems depend on a significant amount and high quality of the dataset, while in the "small data" regime, less work and related applications have been established. The cost of microfluidic experiments and related fluid dynamics simulations limits the scale, predictability, and use in physics-conforming control systems for mobile and low-powered devices. On one hand, the feature extraction from data including images and sequential information in microfluidics is hard due to the potential noisy and limited scale. On the other hand, deep-learning-based simulations mostly rely on labeled training data. This dissertation presents innovative techniques based on data-driven approaches for better solutions in various experimental and computational applications. The first part of this dissertation presents an innovative technique for the detection of rare cells using a deep learning model on microscopic cell images. This proof-of-concept application demonstrates the potential for characterizing rare cells based on image data. Building on this success, we extend the application to an image-based cell sorting system that can identify distinct subpopulations of hematopoietic stem cells and multipotent progenitors. We also delve into the sensitivity of the deep learning-based classifier to the scale of the training dataset. In parallel, we tackle the problem of distortion correction in laser-based galvanometer fabrication systems. Leveraging a hybrid methodology that integrates computer vision algorithms and machine learning, we offer a cloud-based service that enables calibration and correction of lithography systems based on a set of calibration photos. Lastly, we establish a physics-informed model using a stacked U-Net architecture trained in a weakly-supervised manner on computationally-efficient, low-dimensional simulations. This model is created to generate numerical solutions for Navier-Stokes equations without the need for computationally expensive simulation results. An additional physics-informed model is developed to assist in the design of deterministic lateral displacement devices, indicating the potential for leveraging deep learning in microfluidic device design optimization. In essence, this dissertation bridges the gap between deep learning and microfluidics, bringing forth innovative solutions for cell characterization, system calibration, and computational problems within the "small data" regime.
533 $a Electronic reproduction. $b Ann Arbor, Mich. : $c ProQuest, $d 2024
538 $a Mode of access: World Wide Web
650 4 $a Engineering. $3 561152
650 4 $a Mechanical engineering. $3 557493
653 $a Small data
653 $a Deep-learning
653 $a Microfluidics
653 $a Machine learning
653 $a Fluid dynamics
655 7 $a Electronic books. $2 local $3 554714
690 $a 0537
690 $a 0800
690 $a 0548
710 2 $a ProQuest Information and Learning Co. $3 1178819
710 2 $a Lehigh University. $b Mechanical Engineering. $3 845623
773 0 $t Dissertations Abstracts International $g 85-07B.
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30635459 $z click for full text (PQDT)