Deep learning-based molecular morphometrics for kidney biopsies.

Area of Science:

• Nephrology
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Histopathological diagnosis relies on tissue biopsies, but cellular quantification is difficult.
• Podocyte loss in kidney biopsies indicates glomerular injury and disease progression.
• Accurate podocyte counting is crucial for diagnosing and managing kidney diseases.

Purpose of the Study:

• To develop a deep learning approach for antigen-specific cellular morphometrics in human kidney biopsies.
• To accurately quantify podocyte loss and identify morphometric signatures in kidney disease.
• To assess the potential of this method for patient classification and risk stratification in antineutrophil cytoplasmic antibody-associated glomerulonephritis (ANCA-GN).

Main Methods:

• Utilized indirect immunofluorescence imaging combined with U-Net-based deep learning architectures.
• Implemented image-to-image translation and dual segmentation for cellular analysis.
• Analyzed over 27,000 podocytes from 110 human kidney biopsy samples, including ANCA-GN patients.

Main Results:

• Achieved human-level accuracy in cellular morphometrics using the deep learning approach.
• Identified novel morphometric signatures of podocyte depletion in ANCA-GN patients.
• Demonstrated potential for patient classification and risk stratification when combined with clinical tools.

Conclusions:

• The deep learning method enables robust and scalable molecular morphometric analysis of human kidney tissues.
• This approach provides deeper biological insights into kidney pathophysiology, particularly in ANCA-GN.
• The findings support the use of advanced computational methods for improved histopathological diagnosis and patient management.