Development of Method for Quantitative Assessment of Digital Pathology in Woodchuck HCC Model Using Image Segmentation and Image Processing

Authors: Joshua Cheuk, Joshua Cheuk, Nicole Varble, Ming Li, Andrew S. Mikhail, William F. Pritchard, John W. Karanian, Laetitia Saccenti, Bradford J. Wood

Field of Study: Science, Technology, Engineering, and/or Math

Program Affiliation: First-year Research Immersion (FRI), NIH Biomedical Engineering Summer Internship Program

Faculty Mentors: Deborah Kreiss

Easel: 77

Timeslot: Afternoon

Abstract: Hepatocellular carcinoma (HCC) is a leading cause of cancer, diagnosed via imaging (e.g., CT) and histopathology. While CT is minimally invasive, it lacks cellular accuracy, whereas histopathology requires biopsies. This study aims to bridge this gap by standardizing tumor analysis in H&E slides using machine learning. Histopathology samples from woodchucks with HCC were sectioned, stained, digitized, and processed into 256×256 pixel tiles. Tiles were categorized, normalized, and split into hematoxylin (H) and eosin (E) images to enhance cell structure differentiation. The StarDist segmentation model analyzed nuclei, comparing automated and manual counts. Substantially covered tiles (SCT) were the most accurate, with an 91.7 % accuracy, while fully covered blood vessel tiles (FCT-BV) were the least accurate (52.12% accuracy). StarDist struggled with endothelial cells, possibly due to normalization conflicts. Future work aims to refine segmentation models and correlate histopathology with CT imaging for improved, minimally invasive cancer diagnostics.