Date of Award

Spring 5-15-2023

Author's School

McKelvey School of Engineering

Author's Department

Biomedical Engineering

Degree Name

Doctor of Philosophy (PhD)

Degree Type



Optical imaging uses non-ionizing radiation to provide information about the optical absorption and scattering of biological tissues, which are associated with tumor development and progression. This dissertation focuses on algorithmic and machine-learning-based approaches that facilitate the clinical adoption of optical imaging in cancer assessment.Breast cancer is the most diagnosed cancer among women in the United States. However, traditional diagnostic tools, such as X-ray mammography, ultrasound (US), and magnetic resonance imaging (MRI) have low specificity, resulting in numerous unnecessary benign biopsies. Diffuse optical tomography (DOT) has been explored as a potential solution to assist in breast cancer diagnosis and reduce the number of benign biopsies. In the first part of this dissertation, we develop algorithms to correct optode coupling errors and reduce image artifacts in DOT clinical applications. A deep learning-based data generation method is then presented to simplify data acquisition and processing and mitigate mismatch problems in DOT difference imaging. Further, a multi-modality classification strategy with deep learning is described, which enables rapid and accurate classification of breast lesions without the need for BI-RADS readings or hand-crafted features. The second part of the dissertation focuses on colorectal cancer, which is the third most common malignancy and the third leading cause of cancer mortality in the United States. Current endoluminal screening or surveillance for colorectal malignancy is performed by visual endoscopy, which relies on human assessment and can evaluate only the endoluminal surface of the bowel. Optical coherence tomography (OCT) and spatial frequency domain imaging (SFDI) overcome the limitations of traditional camera endoscopy in the gastrointestinal tract, and in this dissertation, a convolutional neural network and AdaBoost classifiers are used to classify OCT and SFDI colorectal images, respectively. Both approaches achieved high accuracy in classifying ex vivo human colorectal tissues. Ovarian cancer is the deadliest of all gynecologic malignancies, and due to the lack of reliable early-stage diagnostic tools, most cases are diagnosed at late stages with a low survival rate. OCT has demonstrated promising results in identifying diseased ovaries and fallopian tubes. In the final part of this dissertation, we calculate the pixel-wise attenuation coefficients of ovaries and fallopian tubes from OCT images and find that statistical features differ significantly between cancerous ovaries, infundibula, and fimbriae and normal ones.


English (en)


Quing Zhu

Committee Members

Adam Bauer, Abhinav Jha, Umberto Villa, Chao Zhou,

Available for download on Thursday, April 18, 2024