Evaluation of Kidney Histological Images Using Unsupervised Deep Learning

Area of Science:

• Nephrology
• Computational Pathology
• Medical Imaging Analysis

Background:

• Machine learning, particularly supervised learning, is increasingly used in medical histopathology.
• Unsupervised learning offers reproducible labeling of histological images but its clinical relevance in nephrology is unclear.
• This study addresses the gap in understanding the relationship between unsupervised histological image analysis and clinical data in kidney disease.

Purpose of the Study:

• To develop and validate an unsupervised approach using convolutional neural networks (CNNs) and visualization for clustering kidney biopsy images.
• To assess the correlation between the derived histological scores and key clinical variables in patients with immunoglobulin A nephropathy.
• To enhance the interpretability and clinical utility of automated histopathological evaluation.

Main Methods:

• An unsupervised method combining CNNs and a visualization algorithm was proposed.
• The approach was applied to hematoxylin and eosin-stained kidney biopsy samples from 68 patients with immunoglobulin A nephropathy.
• Histological scores were calculated and correlated with clinical data including urinary occult blood, urinary protein, serum creatinine, systolic blood pressure, and age.

Main Results:

• The unsupervised method successfully clustered glomeruli into 12 classes and 10 patches.
• The derived histological scores demonstrated significant relationships with assessed clinical variables.
• Specific findings included a significant correlation between serum creatinine and scores from patch-based clusters containing crescentic glomeruli (coefficient = 0.09, P = 0.019).

Conclusions:

• The proposed unsupervised approach effectively extracts clinically relevant features from kidney biopsy images.
• The integrated visualization enhances the interpretability of the findings.
• This method holds potential for the quantified evaluation of renal histopathology and aiding clinical decision-making.