Paper 13305-91
Paper 13305-91
Semantic segmentation and classification of OCT colorectal polyp images
26 January 2025 • 5:30 PM - 7:00 PM PST | Moscone West, Room 2003 (Level 2)
Abstract
Colorectal cancer is the second leading cause of cancer-related deaths, highlighting the need for effective screening methods like colonoscopy. Traditional histopathologic analysis of biopsy samples is effective but costly and time-consuming. To alleviate this burden, it is crucial to accurately evaluate polyps in vivo, distinguishing benign from malignant ones. This study proposes a novel methodology using deep learning algorithms to segment and classify polyps from en face ex vivo Optical Coherence Tomography (OCT) images. A DC-UNet-based model classifies samples as benign or with malignant potential, using a dataset of 143 OCT images annotated by a histologist. The model achieved a training accuracy of 98.08%, with preliminary results indicating OCT's potential to improve early detection of colon malignancies and support a leave-in situ approach, reducing colonoscopy-related burdens.
Presenter
Gabrielle Miller
Texas A&M Univ. (United States)
Gabrielle Miller is an undergraduate Biomechanical and Materials Engineer at Texas A&M University. Her research is conducted through the University of Cyprus KIOS Lab and focuses on machine learning and its applications to the medical field. In addition to her research with KIOS, during the 2024-2025 school year, she will be working in the Biomechanical Environments Laboratory at Texas A&M. She will focus on entrepreneurship and innovation in medical device design while continuing her research in machine learning. Outside of research, she is a TA for calculus and a founding member of the Doctors Without Borders chapter at A&M. She is also a part of the Craig and Galen Honors Engineering College and was awarded the Deans Honor Roll for the College of Engineering. Gabrielle is an active member of her department and was asked to head the programs society, Interdisciplinary Engineering Association.