This study developed and validated an artificial intelligence (AI)-assisted diagnostic model for automated diagnosis of glomerular nephritis (GN) based on renal biopsies, demonstrating high accuracy and stability. Through multicenter data validation, the article highlights the AI model's robust performance in classifying four GN types (IgA nephropathy, membranous nephropathy, focal segmental glomerulosclerosis, and minimal change disease), providing a critical tool to enhance pathologists' diagnostic efficiency and consistency.
Literature Overview
The 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 utilizing 106,988 glomerular microscopic images, the study constructed an integrated model comprising glomerular localization, lesion feature extraction, and patient-level classification, achieving overall performance metrics with F1-scores exceeding 85%. The research further explores dual-branch AI models combining light microscopy with immunofluorescence data to improve diagnostic precision.
Background Knowledge
Glomerular nephritis (GN) represents a group of diseases characterized by glomerular inflammation, serving as a primary cause of end-stage kidney disease, particularly in developing countries. Current GN diagnosis relies on histopathological analysis of renal biopsies, requiring manual identification of glomerular lesions by experienced pathologists, yet this process suffers from subjectivity, time consumption, and poor reproducibility. While AI applications in digital pathology analysis have demonstrated exceptional capabilities in breast and lung cancer diagnostics, standardized and efficient frameworks for GN diagnosis remain lacking. This study pioneers AI implementation in GN pathology diagnostics by constructing an automated, reproducible tool addressing manual slide evaluation limitations. Although previous studies explored AI potential in glomerular segmentation and lesion detection, they lacked large-scale multicenter validation. Through training on over 100,000 images, this research confirms AI model stability in real-world datasets while identifying limitations related to race, disease subtypes, and multiple staining methods, providing directions for future research.
Research Methods and Experiments
This multicenter study analyzed pathological images from 6,682 renal biopsy patients, with 1,235 cases for training, 312 for internal validation, and 2,483/2,652 cases for two external validation cohorts. After quality control and standardization, only Periodic Acid-Silver Methenamine (PASM)-stained light microscopy images were retained. The AI model incorporates three core modules: glomerular localization module (GloSNet) for image segmentation, feature fusion module for lesion feature extraction/integration, and patient-level classification module for final diagnosis. Evaluation metrics included F1-score, precision, recall, and accuracy. Additionally, a dual-branch model integrating immunofluorescence (IF) images was explored to enhance diagnostic accuracy.
Key Conclusions and Perspectives
Research Significance and Prospects
This study demonstrates AI's high accuracy in glomerular nephritis diagnosis for the first time, offering an objective, reproducible tool for pathologists. Future research should expand to additional races and GN subtypes while optimizing model explainability to enhance clinical applicability. The team recommends integrating multiple staining methods and multi-omics data to improve model generalization, enabling broader applications across global medical centers.
Conclusion
This study successfully developed an AI-assisted diagnostic model for glomerular nephritis classification using renal biopsy light microscopy images, achieving efficient differentiation of four GN categories (IgAN, MN, FSGS, MCD). The model demonstrated excellent performance in both external validation cohorts with F1-scores exceeding 83%. Visual analysis confirmed the AI's ability to identify GN-characteristic pathological features. While current limitations exist in race and subtype adaptability, this tool significantly reduces pathologists' workload while improving diagnostic consistency. Future developments include expanding subtypes, integrating multiple staining techniques, enhancing explainability, and exploring real-time diagnostic and telemedicine applications.
