Abstract: FR-PO0694
Evaluating Pediatric Glomerular Basement Membrane Maturation Through Artificial Intelligence Pathomics
Session Information
- Pediatric Nephrology: CKD, ESKD, and Glomerular Diseases
November 07, 2025 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Pediatric Nephrology
- 1900 Pediatric Nephrology
Authors
- Zhou, Zhiqi, Shanghai Jiao Tong University School of Medicine, Shanghai, China
- Guo, Junlin, Vanderbilt University, Nashville, Tennessee, United States
- Wang, Yu, Vanderbilt University Medical Center, Nashville, Tennessee, United States
- Fogo, Agnes B., Vanderbilt University Medical Center, Nashville, Tennessee, United States
- Zhao, Shilin, Vanderbilt University Medical Center, Nashville, Tennessee, United States
- Hou, Tao Xiao, Guangzhou KingMed Center for Clinical Laboratory, Guangzhou, Guangdong, China
- Wang, Xiaowen, Wuhan Children's Hospital, Tongji Medical College, Huazhong Universityof Science & Technology, Wuhan, Hubei, China
- Feng, Chunyue, Zhejiang University School of Medicine Children's Hospital, Hangzhou, Zhejiang, China
- Huo, Yuankai, Vanderbilt University, Nashville, Tennessee, United States
- Yang, Haichun, Vanderbilt University Medical Center, Nashville, Tennessee, United States
- Wu, Ying, Shanghai Jiao Tong University School of Medicine, Shanghai, China
Background
During human development, the glomerular basement membrane (GBM) undergoes neogenesis, regression, and maturation. Previous manual analyses have primarily focused on changes in GBM thickness over time from childhood to adulthood. This study aims to develop an artificial intelligence (AI)-driven pathomics approach to comprehensively evaluate GBM morphology in pediatric patients.
Methods
We analyzed 2,677 electron microscopy (EM) images from 1-2 glomeruli each from 262 Chinese pediatric patients (age range 1-16 yeas) with isolated proteinuria and biopsy-confirmed minimal change disease. GBM thickness was manually measured using the orthogonal intercept method, and an AI-based segmentation model was trained to analyze GBM morphology. Our pathomics analysis included assessments of GBM thickness, curvature, granularity, and six texture features.
Results
The U-Net based AI algorithm effectively segmented the GBM, achieving a DICE score of 0.856. AI-derived GBM thickness measurements correlated well with manual assessments (R=0.72, p<0.001, Fig A). We observed an increase in GBM thickness and a decrease in curvature with age (p<0.001, Fig B), particularly showing linear changes from age 1 to 9 years. Granularity and three texture features—homogeneity, correlation, and energy—increased, while two texture features—dissimilarity and contrast—decreased linearly from ages 1 to 16 years (p<0.05).
Conclusion
We developed an AI-driven pathomics workflow to evaluate GBM morphology during human development, highlighting significant GBM maturation before age 9, with continued subtle changes extending to age 16.
A. The correlation between AI and manual measurement of GBM thickness. B. GBM thickness increase and curvature decrease with age.