Electronic Theses and Dissertations

Date

2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Chemistry

Committee Chair

Xiaohua Huang

Committee Member

Daniel Nascimento

Committee Member

Michael Brown

Committee Member

Qianyi Cheng

Committee Member

Yongmei Wang

Abstract

Single vesicle molecular profiling of cancer-associated EVs is increasingly recognized as a powerful tool for cancer detection and monitoring. Mask and target dual imaging offers a straightforward method to quantify the fraction of molecularly targeted EVs in biofluids at the single vesicle level. However, accurate and efficient dual imaging analysis remains challenging due to interference from false signals in mask images and the need to process large volumes of clinical sample data. A fully automated dual imaging analysis method based on machine learning, integrated with dual imaging single-vesicle technology (DISVT), for breast cancer detection at various stages is presented here. To enhance accuracy and efficiency, we employed the convolutional neural network ResNet34 with transfer learning and developed a machine learning model capable of reliably identifying regions of interest in experimental data. The model was trained using a combination of experimental and synthetic data to optimize its performance. Using DISVT in conjunction with our AI-assisted image analysis platform, we quantified the fractions of EV subpopulations targeting cancer-associated surface protein markers such as EpCAM and CD24 in blood plasma samples from pilot HER2-positive breast cancer patients and healthy donors. Results showed negligible amounts of EpCAM-positive and CD24-positive EVs in both healthy donors and Stage I patients. However, as cancer progressed from Stage II to III, the percentage of EpCAM-positive EVs (also CD81-positive) increased from 18% to 29%, with no significant change observed between Stages III and IV. A similar trend was noted for CD24- positive EVs. Statistical analysis confirmed that both EpCAM and CD24 markers effectively detect HER2-positive breast cancer at Stages II, III, and IV and can differentiate between individual stages, except between Stage III and IV. Due to its simplicity, high sensitivity, and efficiency, DISVT combined with AI-driven dual imaging analysis holds significant potential for both fundamental research and clinical applications, enabling precise molecular characterization of targeted EV subtypes in biofluids.

Comments

Data is provided by the student.

Library Comment

Dissertation or thesis originally submitted to ProQuest.

Notes

Open access

Download

Share

COinS