Deep Learning-Based Segmentation and Quantification in Experimental Kidney Histopathology.

Area of Science:

• Nephrology
• Computational Pathology
• Biomedical Imaging

Background:

• Kidney disease research relies on histopathology for understanding pathophysiology.
• Precision medicine necessitates quantitative, reproducible, and efficient histopathology tools.
• Deep learning, specifically convolutional neural networks (CNNs), shows promise for automated histology segmentation.

Purpose of the Study:

• To develop and validate a CNN for accurate segmentation of kidney structures in various animal models.
• To enable high-throughput, quantitative histopathologic analysis for kidney disease research.

Main Methods:

• A CNN architecture was trained to segment six key renal structures in periodic acid-Schiff-stained kidney tissue.
• Expert-based annotations (72,722) were used to train the CNN for high accuracy.
• The CNN was tested on healthy mice, five murine disease models, and tissues from other species, including humans.

Main Results:

• The CNN achieved very high multiclass segmentation performance across all disease models.
• Quantitative analysis revealed interstitial expansion, tubular dilation/atrophy, and glomerular size variability in disease models.
• The CNN demonstrated high performance in rats, pigs, bears, marmosets, and humans, showing translational potential.

Conclusions:

• A deep learning algorithm was developed for accurate segmentation of kidney digital whole-slide images.
• This tool enables reproducible, quantitative histopathologic analysis in preclinical models.
• The findings suggest applicability to clinical studies, bridging preclinical and clinical research.