Dissertations and Theses

Date of Award


Document Type




First Advisor

Yingli Tian


Computer vision, Deep learning, Medical image analysis, Artificial intelligence, Computer-aided diagnosis


Medical imaging has been applied widely in many clinical diagnoses to detect and differentiate abnormalities by revealing the internal structure of the human body at normal anatomical and physiological levels. Manual analyzing medical images demands attention and is time-consuming, requiring well-trained expertise. The speed, fatigue, and experience may limit the diagnostic performance, leading to delays and even false diagnoses that significantly impact patient treatment. Therefore, accurate systematic systems based on medical image analysis are crucial for timely clinical diagnosis.

This dissertation focuses on advancing automatic computer-aided diagnosis systems to detect cancer, assisting radiologists with early intervention to improve survival rates. Many researchers have employed Artificial intelligence (AI) in automatic diagnosis systems for cancer detection and assessment. However, existing work is still limited to safer medical imaging modalities, accurate automated cancer diagnosis, efficient image annotation acquisition, and robustness in different domains.

This dissertation aims to address the remaining concerns within the following four aspects: (1) Contrast-enhanced medical images for computer-aided diagnosis. It can effectively highlight the internal structure of blood vessels and organs, help abnormal detection, and is a better image representation. (2) Effective medical image annotation acquisition. It significantly saves the expensive annotation costs, accelerates the training process, and prevents AI model overfitting with limited training images. (3) Automatic cancer diagnosis system with high sensitivity and specificity. It is required to accurately detect, segment, and classify abnormality on medical images and can be adapted to multiple manufacturers of medical images. (4) Identifying out-of-distribution data by uncertainty estimation. It prevents the AI-driven diagnostic model from generating over-confident predictions due to image distribution shifts. This dissertation reviews the previous research for these tasks and remaining issues, as well as proposes AI-driven solutions to tackle these issues and presents the results.



To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.