Liu, Chang
(2024)
Machine learning on medical and biological images to support clinical applications.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
Artificial intelligence (AI) has achieved remarkable success in biomedical-related applications. The overall goal of this dissertation is to develop novel AI and machine learning techniques for three biomedical use cases to enhance clinical workflows, provide valuable insights into complex medical conditions, and facilitate the biological manufacturing process. First, in breast cancer diagnosis, the high rate of false-positive biopsy findings is a pressing issue. This leads to unnecessary invasive procedures, increased patient anxiety, and financial costs. We developed AI models to aid radiologists in precise diagnosis towards reducing unnecessary biopsies, enhancing workflow efficiency, and alleviating patient stress. Specifically, we investigated two different breast imaging modalities, digital mammography, and contrast-enhanced mammography (CEM), to develop quantitative characterization, radiomics, and deep learning methods. Second, in post-cardiac arrest brain injury, accurately assessing neurological prognosis for comatose survivors is challenging in critical care practice. We developed deep learning models to investigate the relationship between early brain Computed Tomography (CT) images and Electroencephalogram (EEG) signals, towards identifying patients at risk of developing highly malignant finding to make informed decisions about post-cardiac arrest treatment. We also customized interpretable AI models to identify clinically relevant CT image patterns to facilitate a better visual understating of brain injury post-cardiac arrest, enhancing the trust of critical care physicians in AI model’s output. Third, towards fast manufacturing of human-induced pluripotent stem cells (hiPSC)-derived organoids, we developed machine learning models to facilitate cell/organoid selection through staining-free imaging. This innovative approach allows for non-invasive monitoring of morphological characteristics at different stages of cell development, eliminating the need for destructive staining. AI-enabled cell/organoid fast manufacturing can benefit disease modeling and drug development, as it enables real-time feedback on the state of cell/organoid, especially in high-throughput drug testing. In the three biomedical scenarios, we developed and evaluated multiple machine/deep learning models, along with technical innovation, to address unmet needs and advance biomedical-driven AI for clinical/translational applications.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
| Title | Member | Email Address | Pitt Username | ORCID |
|---|
| Committee Chair | Wu, Shandong | | | | | Committee Member | Banerjee, Ipsita | | | | | Committee Member | Kim, Kang | | | | | Committee Member | Arefan, Dooman | | | |
|
| Date: |
6 March 2024 |
| Defense Date: |
23 April 2024 |
| Approval Date: |
7 January 2025 |
| Submission Date: |
8 November 2024 |
| Access Restriction: |
2 year -- Restrict access to University of Pittsburgh for a period of 2 years. |
| Number of Pages: |
125 |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
Swanson School of Engineering > Bioengineering |
| Degree: |
PhD - Doctor of Philosophy |
| Thesis Type: |
Doctoral Dissertation |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
Artificial intelligence, machine learning, deep learning, breast cancer, cardiac arrest, medical image |
| Date Deposited: |
07 Jan 2025 21:10 |
| Last Modified: |
07 Jan 2025 21:10 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/47067 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[img]](https://d-scholarship.pitt.edu/style/images/fileicons/application_pdf.png)
![[feed]](/images/feed-icon-32x32.png){kind=icon}