Artificial intelligence-assisted diagnosis of glomerular nephritis using a pathological image analysis approach: a...

Area of Science:

• Nephrology
• Pathology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Glomerular nephritis (GN) diagnosis relies on manual kidney biopsy analysis, which is subjective and labor-intensive, leading to interobserver variability.
• Developing an automated diagnostic tool can enhance objectivity and efficiency in GN diagnosis.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI)-assisted model for diagnosing four common types of GN using digital histological images.
• The model aims to segment glomeruli, extract features, and classify diseases including IgA nephropathy (IgAN), membranous nephropathy (MN), minimal change disease (MCD), and focal segmental glomerulosclerosis (FSGS).

Main Methods:

• A multicentre study utilized 106,988 glomeruli images from 6682 patients across three institutions for model development and validation.
• The AI model comprises three modules: glomerular localization (GloSNet), feature extraction/fusion for lesion classification, and patient-level diagnosis.
• Performance was evaluated using F1-score, precision, and recall on training, internal validation, and two external validation cohorts.

Main Results:

• The AI model demonstrated strong performance in external validation cohorts, achieving F1-scores of 83.86% and 85.45%, precision of 81.37% and 83.12%, and recall of 87.84% and 88.94%, respectively.
• The model successfully segmented glomeruli, identified lesions, and classified four types of GN with high accuracy.

Conclusions:

• An AI-assisted model can accurately diagnose four common GN subtypes from kidney biopsy images, offering potential to reduce pathologist workload and improve diagnostic efficiency.
• Further research is needed to address model interpretability, incorporate diverse staining methods, and expand the range of diagnosed GN subtypes.