Correlating Deep Learning-Based Automated Reference Kidney Histomorphometry with Patient Demographics and Creatinine.

Area of Science:

• Nephrology
• Computational Pathology
• Medical Imaging Analysis

Background:

• Establishing reference histomorphometric data for healthy human kidneys is challenging due to complex quantitation.
• Machine learning can correlate histomorphometric features with clinical parameters to understand population variance.
• This study investigates the relationship between kidney histomorphometry and patient age, sex, and serum creatinine (SCr) using deep learning.

Approach:

• A deep learning model performed panoptic segmentation of kidney tissue images to identify and quantify viable/sclerotic glomeruli, interstitia, tubules, and vasculature.
• Morphometric parameters like size and density were extracted from segmented compartments.
• Regression analysis determined correlations between histomorphometric features and patient demographics (age, sex, SCr).

Key Points:

• Deep learning model demonstrated high segmentation accuracy for kidney tissue compartments.
• Significant variations in nephron and vasculature size/density were observed among healthy individuals.
• Nephron size correlated with SCr; renal vasculature showed sex-based differences; glomerulosclerosis and cortical artery density varied with age.

Conclusions:

• Deep learning enables automated, precise measurement of kidney histomorphometric features.
• Histomorphometric features in reference kidney tissue significantly correlate with patient demographics and SCr.
• Deep learning enhances the efficiency and rigor of histomorphometric analysis for kidney research.