eClinicalMedicine | AI-Assisted Diagnosis of Glomerular Nephritis

This study developed and validated an artificial intelligence (AI)-assisted diagnostic model for automated diagnosis of glomerular nephritis (GN) based on kidney biopsies, demonstrating high accuracy and stability. Through multicenter data validation, the paper highlights the AI model's robust performance in diagnosing four GN types (IgA nephropathy, membranous nephropathy, focal segmental glomerulosclerosis, and minimal change disease), offering an essential tool to improve pathologists' efficiency and diagnostic consistency.

 

Literature Overview
This article, 'Artificial intelligence-assisted diagnosis of glomerular nephritis using a pathological image analysis approach: a multicentre model development and validation study', published in eClinicalMedicine, reviews the model development and validation process for AI-assisted glomerular nephritis diagnosis. Through multicenter research, it constructed an integrated model using 106,988 glomerular microscopic images, comprising three components: glomerular localization, lesion feature extraction, and patient-level classification, achieving an overall performance F1-score exceeding 85%. The study also explored a dual-branch AI model combining light microscopy and immunofluorescence data to further enhance diagnostic precision.

Background Knowledge
Glomerular nephritis (GN) is a group of diseases characterized by glomerular inflammation and a major cause of end-stage kidney disease, particularly in developing countries. Currently, GN diagnosis relies on histopathological analysis of kidney biopsies, a process requiring experienced pathologists to manually identify glomerular lesions, yet suffers from subjectivity, time consumption, and poor reproducibility. While AI has shown remarkable capabilities in digital pathology analysis for diseases like breast and lung cancer, a unified and efficient framework for GN diagnosis remains lacking. This study pioneers AI application in GN pathology, creating an automated, reproducible diagnostic tool that addresses manual slide review limitations. Though prior studies explored AI in glomerular segmentation and lesion detection, they lacked large-scale multicenter validation. By training on over 100,000 images, this study confirmed AI model stability in real-world data, identified limitations in race, disease subtypes, and staining methods, and provided directions for future research.

 

 

Research Methods and Experiments
This multicenter study collected pathological images from 6,682 kidney biopsies, with 1,235 cases for training, 312 for internal validation, and 2,483/2,652 for two external validation cohorts. All images underwent quality control and standardization, retaining only Periodic Acid-Silver Methenamine (PASM)-stained light microscopy images. The AI model consists of three core modules: glomerular localization (GloSNet) for image segmentation, feature fusion for extracting and integrating glomerular lesion features, and patient-level classification for final diagnosis. Evaluation metrics included F1-score, precision, recall, and accuracy. Additionally, the study explored a dual-branch model incorporating immunofluorescence (IF) images to enhance diagnostic accuracy.

Key Conclusions and Perspectives

• The AI model achieved F1-scores of 83.86% in external validation cohort 1 and 85.45% in cohort 2, demonstrating high diagnostic consistency across datasets.
• The glomerular segmentation module GloSNet performed exceptionally at both pixel (93.01% F1-score) and glomerular levels (96.10% F1-score), showing high accuracy in automated glomerular region identification.
• The feature fusion strategy, based on kernel density estimation, effectively integrated diverse lesion features, improving multi-category GN classification performance, particularly for membranous nephropathy (MN) with 97.13% F1-score.
• The dual-branch AI model combining immunofluorescence data improved four-class diagnostic F1-score to 91.00%, confirming IF data's enhancement of AI diagnostic capability.
• Through attention heatmap visualization, the AI model demonstrated precise identification of GN-associated pathological features, such as MN basement membrane thickening and FSGS sclerosis regions.

Research Significance and Prospects
This study demonstrates AI's high accuracy in GN diagnosis, providing pathologists with an objective, reproducible tool. Future research should expand to more races and GN subtypes while optimizing model interpretability to improve clinical applicability. The team recommends integrating multiple staining methods and multi-omics data to enhance generalization for global clinical application.

 

 

Conclusion
This study successfully established an AI-assisted diagnostic model for glomerular nephritis classification using light microscopy kidney biopsy images, achieving efficient four-class differentiation (IgAN, MN, FSGS, MCD). The model demonstrated superior performance across two external validation cohorts with F1-scores exceeding 83%. Visualization analysis confirmed the AI's ability to identify GN-associated pathological features. While current limitations exist in race and subtype adaptability, this tool significantly reduces pathologists' workload and improves diagnostic consistency. Future directions include expanding to additional subtypes, integrating multiple staining protocols, enhancing model interpretability, and exploring real-time diagnostic and telemedicine applications.

 

Sheng Nie, Nan Jia, Haobo Chen, Qi Zhang, and Fan Fan Hou. Artificial intelligence-assisted diagnosis of glomerular nephritis using a pathological image analysis approach: a multicentre model development and validation study. eClinicalMedicine.