Computer Science Department, MS Thesis Presentation, Jessica Elmhurst " Automated Scan-Viability Mapping Using Image Segmentation for Optical Coherence Tomography on Deceased Donor Kidneys"

Jessica Elmhurst 

MS Student

WPI – Computer Science Department 

Tuesday, April 28^th, 2026 

Time: 10:00am-11:00am

Location: Unity Hall 446 Tech Suite

Advisor: Prof. Haichong K. Zhang

Reader: Prof. Bahman Moraffah

Abstract:

Current deceased donor kidney assessment often leads to the disqualification of viable organs. Optical Coherence Tomography (OCT) is a powerful and non-invasive imaging modality that produces high-resolution images of tissue microstructures associated with post-transplant renal function. Integrated with a robotic system, the process of imaging renal microstructures with OCT has become faster and capable of capturing spatially dense information with large-area scans. However, the presence of fat tissue on the surface of the kidney can attenuate the OCT scan signal and delay scan acquisition. To resolve this, we present a deep learning pipeline to identify scan-viable regions for precise robotic tissue targeting.

Using a two-tiered U-Net architecture trained on a dataset of deceased donor kidneys, the pipeline first isolates the kidney contour and then segments adipose tissue.  The resulting inference masks are used to create a scan viability map and virtual trajectories to assist robotic probe movement and optimize scan acquisition. Validated against state-of-the-art models using Dice Coefficient and IoU metrics, the pipeline will augment pre-transplant organ evaluation and enhance organ assessment workflows.

Thank you so much for your guidance with this process. Please let me know if you have any questions or something I should change.