Improved Glomerular Nephritis Diagnosis With Artificial Intelligence-Assisted Model

A new methodology leveraging artificial intelligence to standardize the diagnosis of glomerular nephritis (GN) could revolutionize kidney biopsy analysis.1

The dual-branch, multicenter study included data from more than 6600 patients across three hospitals and diagnostic centers in China. This is the first research study to utilize an AI-assisted model for GN diagnosis and has shown an improvement in diagnostic standardization, efficiency, and the removal of interobserver variability.2

Investigators suggest an automated artificial intelligence-assisted model is an effective tool to remove these inconsistencies and effectively distinguish between focal segmental glomerulosclerosis (FSGS), IgA nephropathy (IgAN), membranous nephropathy (MN), or minimal change disease (MCD).

“The diagnosis of GN heavily relies on the analysis of histopathological images obtained from kidney biopsies,” wrote Fan Fan Hou, MD, chief of the renal division at the Nanfang Hospital in Guangzhou, China, and colleagues. “This process involves manual interpretation by experienced pathologists, who assess glomeruli's morphological, structural, and compositional changes. However, this approach is subjective, time-consuming, and labor-intensive, which limits the accuracy and reproducibility of diagnosis.”

The AI-assisted diagnostic model utilized 106,988 glomeruli light microscopy images to classify glomerular lesions and employed immunofluorescence (IF) to interpret immune marker patterns and extract complementary diagnostic features. Performance of the model was evaluated using F1-score, precision, recall, and accuracy.

To develop the AI model, researchers performed computational analysis of 6682 retrospective and deidentified patient records from 3 medical centers. The model’s training and internal validation cohorts included 1235 and 312 patient groups, respectively, from the Nanfang Hospital of Southern Medical University. The external validation cohorts I and II included 2484 patients from the Jinyu Diagnostic Center (I) and 2652 from the Huayin Diagnostic Center (II).

In the internal validation cohort, the different types of GN diagnoses were represented by 72 patients with FSGS (23.08%), 81 patients with IgAN (25.96%), 80 patients with MCD (25.64%), and 79 patients with MN (25.32%). In external validation cohort I, there were 110 patients with FSGS (4.43%), 914 patients with IgAN (36.81%), 392 patients with MCD (15.79%), and 1067 patients with MN (42.97%). In the external validation cohort II, there were 183 patients with FSGS (6.90%), 891 patients with IgAN (33.60%), 222 patients with MCD (8.37%), and 1356 patients with MN (51.13%).

From the results of the AI-assisted model’s data, the internal validation cohort showed scores of F1 at 84.48%, precision at 85.48%, and recall at 85.52%. For external validation cohort I, F1 was 83.86% (95% confidence interval CI, 81.64–86.03, precision was 83.86% (95% confidence interval CI 81.64–86.0), and recall was 87.84% (95% CI 85.49–89.95). In external validation cohort II, the F1-score was 85.45% (95% CI 83.66–87.29, the precision was 83.12% (95% CI 81.26–85.03), and the recall was 88.94% (95% CI 87.07–90.79).

Further analysis of the findings shows the AI-assisted light microscopy-based GND model was efficient in classifying the specified four types of GN. Specifically, in the external validation cohort I, MN demonstrated the highest F1-score with a significant rate of 97.13% (95% CI 96.33–97.76), meaning the model exhibited outstanding levels of precision and recall. In external validation cohort II, the F1-score for MCD was 81.08% (95% CI 77.09–84.68) and showed the model’s difficulty with distinguishing between MCD and FSGS, a very common clinician error.

“To our knowledge, it is the first study to develop an AI-assisted model to diagnose GN with images of kidney biopsy tissues. Second, the sample size is large enough to develop and externally validate the AI-assisted model,” researchers concluded. “However, there were also several limitations in our study. Although we had externally validated the AI-assisted model with independent datasets, if the model can be used in other populations with different racial or ethnic backgrounds needed further validation.”

1. Nie S, Jia N, Chen H, et al. Artificial intelligence-assisted diagnosis of glomerular nephritis using a pathological image analysis approach: a multicentre model development and validation study. eClinicalMedicine. 2025;89:103530. :https://doi.org/10.1016/j.eclinm.2025.103530

2. Gonzalez FM, Valjalo R. Essential role of kidney biopsy in diagnosing glomerular diseases amidst evolving biomarkers. World Journal of Nephrology. 2025;14(2). :https://doi.org/10.5527/wjn.v14.i2.103756‌