GNCnn: A QuPath extension for glomerulosclerosis and glomerulonephritis characterization based on deep learning.

Area of Science:

• Digital pathology
• Bioimage analysis
• Artificial intelligence in medicine

Background:

• Digitalization of microscopy enables AI applications in pathology.
• QuPath is a widely used bioimage analysis software.
• Nephropathology requires specialized tools for analyzing kidney structures.

Purpose of the Study:

• To introduce GNCnn, the first open-source QuPath extension for nephropathology.
• To provide an automated tool for detecting and classifying glomeruli using deep learning.
• To support nephropathologists in diagnosing kidney diseases like glomerulosclerosis and glomerulonephritis.

Main Methods:

• Development of GNCnn as a QuPath extension.
• Integration of deep learning models for glomeruli detection and classification.
• User-friendly interface for real-time analysis at glomerulus and slide levels.

Main Results:

• High accuracy in glomeruli detection (Dice coefficient: 0.807).
• Accurate classification of glomeruli as sclerotic or non-sclerotic (balanced accuracy: 98.46%).
• Effective classification of non-sclerotic glomeruli into 12 types of glomerulonephritis (top-3 balanced accuracy: 84.41%).

Conclusions:

• GNCnn offers an accessible, automated solution for nephropathology analysis.
• The tool streamlines the diagnostic workflow within the QuPath environment.
• GNCnn accelerates and supports the diagnosis of kidney diseases by integrating analysis into the pathologist's workspace.