Aishwarya Arvind
Pronouns: She/Her
Research Mentor(s): Lucas Junginger
Research Mentor School/College/Department: Orthopaedic Surgery / Medicine
Program: UROP
Session: Session 4 (1:40pm – 2:30pm)
Authors: Aishwarya Arvind, Lucas Junginger
Abstract
Introduction: In osteoarthritis research, tissue histology is used to evaluate disease progression and severity. However, qualitative grading of images is time-consuming and a source of inter-operator variability. Here, we demonstrate an automated approach to disease grading using Convolutional Neural Networks to segment blood vessels within images of synovial tissue. Methods: Synovial explants were obtained from arthritic patients undergoing knee replacement, then processed for histological sectioning, stained with hematoxylin and eosin (H&E), and imaged. Training masks were drawn to identify any blood vessels from the surrounding background region, broken down into three classes: background, tissue, vessel. The dataset was partitioned into training and validation sets (n=30/10), then used to train a CNN. Segmentation results were then post-processed using standard image analysis techniques. All analysis was done using MATLAB 2022a and its “Statistics and Machine Learning” toolbox. The network architecture was a standard U-Net. Results: Segmentation accuracy was assessed by comparison to the validation set. With regards to the background vs tissue segmentation, a validation accuracy of 97%. Discussion/Clinical Significance: The ultimate goal of this project is to segment each of the features of the tissue images, and develop an automated grading system to determine the severity of the osteoarthritis present in tissue.
