Animal Internal Motion Analysis with Unsupervised Machine Learning Methods

dc.contributor.authorZheng, Weien
dc.contributor.committeechairWang, Yue J.en
dc.contributor.committeechairYu, Guoqiangen
dc.contributor.committeememberLin, Zinen
dc.contributor.committeememberZhang, Richarden
dc.contributor.committeememberJi, Boen
dc.contributor.departmentElectrical and Computer Engineeringen
dc.date.accessioned2025-05-31T08:03:29Zen
dc.date.available2025-05-31T08:03:29Zen
dc.date.issued2025-05-30en
dc.description.abstractUnderstanding the complex internal motions within biological systems—from cellular migrations during development to repeated contractions in muscle tissue—is essential for comprehending the fundamental mechanisms that drive life processes. This report presents innovative unsupervised machine learning methods to explore the dynamics of cellular and tissue motion and their implications. We first address cellular motion, focusing on long-term movements. While our laboratory's previously established Minimum-Cost Circulation-Based Framework provides a foundational approach, it suffers from limitations in robustness due to inadequate accuracy in cell segmentation. To address these shortcomings, we introduce PrinCut-Auto, a cutting-edge method leveraging Multi-scale Principal Curvature (MSPC) for precise cell identification, integrated with min-cut optimization and order-statistics testing to significantly advance 3D cell segmentation technology. PrinCut-Auto can address the limitations in the accuracy and robustness of the framework and we provide comprehensive experimental evidence of its efficacy. Beyond cellular movements, the study also addresses the analysis of tissue dynamics, with a particular emphasis on rapid and repetitive motions such as muscle contractions. To capture these complex patterns effectively, we developed BWMquant, a tool capable of efficient and accurate motion correction, as well as the detection of recurring motion patterns within tissues. The impact of these tissue dynamics on biological processes is discussed, with specific experimental setups and results demonstrating the crucial interplay between mechanical movements and biological functionality. The report concludes with a discussion on future directions, emphasizing the potential for these methodologies to enhance biological research. This study not only advances our understanding of biological internal motion but also sets the stage for future innovations in the field.en
dc.description.abstractgeneralUnderstanding the intricate movements within living organisms—ranging from the migration of individual cells during development to the contractions of muscle tissues—is fundamental to comprehending life's essential processes. Recent advancements in imaging technologies, such as three-dimensional (3D) light-sheet microscopy, have enabled scientists to observe these internal motions in unprecedented detail. However, the sheer volume and complexity of this image data present a serious challenge. Analyzing these datasets manually is slow, error-prone, and often infeasible. This dissertation was motivated by the need for automated tools that can make sense of these intricate biological processes. Specifically, it introduces unsupervised machine learning methods that require no manually labeled data, making them well-suited for handling the vast and diverse image data generated in modern biological research. Focusing first on cellular motion, particularly the long-term tracking of cells during developmental processes, previous methodologies have faced limitations in robustness, primarily due to challenges in accurately segmenting individual cells from complex 3D images. To overcome these obstacles, we developed PrinCut-Auto, a state-of-the-art method that enhances 3D cell segmentation technology. PrinCut-Auto employs Multi-scale Principal Curvature (MSPC) analysis to precisely identify cell boundaries, even in densely packed cellular environments. This technique is integrated with min-cut optimization and order-statistics testing, allowing for the automatic and accurate delineation of individual cells without the need for manually labeled training data. Extensive experimental evaluations demonstrate that PrinCut-Auto significantly improves the accuracy and robustness of cell segmentation, thereby enhancing the reliability of subsequent cell tracking analyses. The second core tool developed in this dissertation is BWMquant, a method for correcting and interpreting motion in living tissue. Living animal imaging often suffers from highly non-rigid motion in the sample, which can obscure biological signals. BWMquant automatically stabilizes image sequences by correcting these drifts. More importantly, it also detects repeating patterns of movement within the tissue itself—such as rhythmic contractions, pulsations, or other repeating dynamics, facilitating a deeper understanding of how these dynamics influence overall biological function. Through specific experimental setups, our method helps biologists to explore the critical interplay between mechanical movements and biological processes, highlighting how tissue motion contributes to development and function. Together, these tools represent a step forward in the automation and scalability of biological image analysis. Their performance exceeds that of many current supervised and unsupervised approaches, and they are designed with usability in mind, so that biologists without coding experience can readily apply them. By reducing the technical barriers to analyzing complex image data, this work empowers researchers to explore new scientific questions about how life builds itself from within.en
dc.description.degreeDoctor of Philosophyen
dc.format.mediumETDen
dc.identifier.othervt_gsexam:43106en
dc.identifier.urihttps://hdl.handle.net/10919/134955en
dc.language.isoenen
dc.publisherVirginia Techen
dc.rightsCreative Commons Attribution 4.0 Internationalen
dc.rights.urihttp://creativecommons.org/licenses/by/4.0/en
dc.subjectoptimizationen
dc.subjectstatisticsen
dc.subjectmachine learningen
dc.subjectimage segmentationen
dc.subjectimage registrationen
dc.titleAnimal Internal Motion Analysis with Unsupervised Machine Learning Methodsen
dc.typeDissertationen
thesis.degree.disciplineComputer Engineeringen
thesis.degree.grantorVirginia Polytechnic Institute and State Universityen
thesis.degree.leveldoctoralen
thesis.degree.nameDoctor of Philosophyen

Files

Original bundle
Now showing 1 - 1 of 1
Loading...
Thumbnail Image
Name:
Zheng_W_D_2025.pdf
Size:
89.65 MB
Format:
Adobe Portable Document Format