Abstract: SA-PO047
Analyzing the Influence of Glomerulus Structural Features Using Minimum Spanning Trees
Session Information
- Engineering-Based Approaches to Problems in Nephrology
November 09, 2019 | Location: Exhibit Hall, Walter E. Washington Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Bioengineering
- 300 Bioengineering
Authors
- Border, Samuel P., SUNY Buffalo, Buffalo, New York, United States
- Jen, Kuang-Yu, University of California, Davis, Sacramento, California, United States
- Jain, Sanjay, Washington University School of Medicine, St. Louis, Missouri, United States
- Fogo, Agnes B., Vanderbilt University Medical Center, Nashville, Tennessee, United States
- Tomaszewski, John E., SUNY Buffalo, Buffalo, New York, United States
- Sarder, Pinaki, SUNY Buffalo, Buffalo, New York, United States
Background
Patients with Type 1 or Type 2 Diabetes mellitus are at an increased risk for extensive vascular dysfunction which leads to Diabetic Nephropathy (DN) in close to 40% of cases. Structural changes within the glomerulus resulting from the diabetic phenotype are used to classify glomeruli into different stages of DN. Using Minimum Spanning Trees (MST’s) we are able to quantitatively define these structural abnormalities and assess their contribution to DN stage of that glomerulus. We employed Bayesian Networks (BN) to visualize and assess the relationships between different MST features and the glomerulus’s DN stage.
Methods
We used stain deconvolution to isolate nuclear regions in 799 H&E stained glomerulus images. The centroids of glomerular nuclei were used as nodes to create a MST in Matlab. Features calculated from these MST’s quantified spread, connectedness, and fraction of terminal “leaf” nodes present as a representation of degree of glomerular expansion, cellularity, and sclerosis. Using these features, a BN was generated in R using the Hill-Climbing structure learning method. The resulting network was thresholded so that only the most significant relationships were included.
Results
Graph features calculated from the glomerular MST had a significant influence on the DN Stage. From this network, we are able to generate conditional probability distributions for each of the feature values that can be used to predict the DN classification.
Conclusion
By analyzing glomeruli using MST’s we can impose repeatable, robust measures of nuclei distribution and derive further information as to what biological factors contribute to DN progression. Incorporation of BN and other probabilistic graphical models in medicine allows for more informed diagnosis and research.
Funding
- NIDDK Support