Research
About
Research Overview: Understanding Type 1 Diabetes Progression through Advanced Imaging Techniques
​
Research Project Title: Enhancing Quantitative Analysis of Pancreatic Islets in Type 1 Diabetes Using Deep Learning
​
Principal Investigator: Dr. Sarah Kim
Institution & Department: College of Pharmacy, University of Florida
Project Duration: March 2024 – Present
​
Research Focus: My research is centered on the progression of Type 1 Diabetes (T1D), a chronic autoimmune condition where the body’s immune system attacks insulin-producing beta cells in the pancreas. Specifically, this project leverages the extensive whole slide imaging (WSI) data from the JDRF Network for Pancreatic Organ Donors with Diabetes (nPOD) to analyze the heterogeneity of pancreatic islets—clusters of cells that include insulin-producing beta cells—across different stages of the disease.
​
In patients with T1D, the islets display significant variability in size, number, and cellular composition when compared to non-diabetic controls. This project aims to enhance the efficiency and accuracy of WSI analysis using deep learning models, enabling a more precise understanding of T1D progression. By utilizing advanced computational approaches, including the Segment Anything Model (SAM) and QuPath pixel classifiers, we can automate the identification and quantification of key cellular markers within the islets. This includes the detailed analysis of immune cell infiltration, particularly CD3+ cells, which play a critical role in the autoimmune response in T1D.
​
Project Responsibilities: Since joining the project in March 2024, my primary responsibility has been training the deep learning models within QuPath, an open-source software platform widely used for WSI analysis. I have focused on optimizing the workflow for analyzing pancreatic tissue samples, using the University of Florida's HiPerGator supercomputer to manage and process large data sets.
​
My work involves applying the SAM model to define the precise boundaries of islets within regions of interest, as well as using the pixel classifier to segment insulin and glucagon-producing areas within each islet. This process is critical for accurately quantifying the presence of CD3+ cells both inside and around the perimeter of the islets, which provides valuable insights into the presence and severity of insulitis—a hallmark of T1D.
The results from our workflow have shown promising accuracy in quantifying these features, paving the way for larger-scale analysis. Ultimately, our goal is to utilize this workflow to analyze a broader set of WSI data from the nPOD program, which will significantly contribute to our understanding of T1D progression and may inform future therapeutic strategies.