A U-Net based framework to quantify glomerulosclerosis in digitized PAS and H&E stained human tissues.

Area of Science:

• Nephrology
• Digital Pathology
• Artificial Intelligence

Background:

• Accurate glomeruli counting and glomerulosclerosis evaluation are crucial for diagnosing kidney diseases.
• Manual microscopic assessment is time-consuming and labor-intensive.

Purpose of the Study:

• To develop and validate a deep learning (DL) approach for automated identification and classification of normal and sclerosed glomeruli in digital whole slide images (WSIs).
• To assess the performance of the DL model across different staining methods (PAS and H&E).

Main Methods:

• A U-Net deep learning model was trained on Periodic Acid-Schiff (PAS) stained WSIs.
• The model segmented and classified glomeruli, with classifications refined by histomorphometry.
• The model was tested on independent PAS and Hematoxylin and Eosin (H&E) stained WSIs from multiple institutions.

Main Results:

• The DL model achieved high F1-scores for classifying normal glomeruli (97.5% in PAS, 94.5% overall) and sclerosed glomeruli (68.8% in PAS, 76.8% overall).
• The model demonstrated reliable performance on both PAS and H&E stained images.
• The developed framework showed higher accuracy than some existing methods.

Conclusions:

• A DL framework based on the U-Net model can reliably segment and classify normal and sclerosed glomeruli in both PAS and H&E stained WSIs.
• This automated method offers a more efficient and accurate alternative to manual assessment for kidney disease evaluation.
• The study provides a publicly available dataset for further research in digital pathology for renal diagnostics.