Date of Graduation

12-2022

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Engineering (PhD)

Degree Level

Graduate

Department

Computer Science & Computer Engineering

Advisor/Mentor

Alexander Nelson

Committee Member

David Andrews

Second Committee Member

James Parkerson

Third Committee Member

John Tipton

Keywords

Artificial intelligence, Breast cancer detection, Deep learning, Image processing, Pulsed terahertz imaging

Abstract

Breast cancer affects about 12.5% of women population in the United States. Surgical operations are often needed post diagnosis. Breast conserving surgery can help remove malignant tumors while maximizing the remaining healthy tissues. Due to lacking effective real-time tumor analysis tools and a unified operation standard, re-excision rate could be higher than 30% among breast conserving surgery patients. This results in significant physical, physiological, and financial burdens to those patients. This work designs deep learning-based segmentation algorithms that detect tissue type in excised tissues using pulsed THz technology. This work evaluates the algorithms for tissue type classification task among freshly excised tumor samples. Freshly excised tumor samples are more challenging than formalin-fixed, paraffin-embedded (FFPE) block sample counterparts due to excessive fluid, image registration difficulties, and lacking trustworthy pixelwise labels of each tissue sample. Additionally, evaluating freshly excised tumor samples has profound meaning of potentially applying pulsed THz scan technology to breast conserving cancer surgery in operating room. Recently, deep learning techniques have been heavily researched since GPU based computation power becomes economical and stronger. This dissertation revisits breast cancer tissue segmentation related problems using pulsed terahertz wave scan technique among murine samples and applies recent deep learning frameworks to enhance the performance in various tasks. This study first performs pixelwise classification on terahertz scans with CNN-based neural networks and time-frequency based feature tensors using wavelet transformation. This study then explores the neural network based semantic segmentation strategy performing on terahertz scans considering spatial information and incorporating noisy label handling with label correction techniques. Additionally, this study performs resolution restoration for visual enhancement on terahertz scans using an unsupervised, generative image-to-image translation methodology. This work also proposes a novel data processing pipeline that trains a semantic segmentation network using only neural generated synthetic terahertz scans. The performance is evaluated using various evaluation metrics among different tasks.

Download

Share

COinS