Dynamic parametric MRI and deep learning: Unveiling renal pathophysiology through accurate kidney size quantification

Area of Science:

• Biomedical Imaging
• Artificial Intelligence in Medicine
• Renal Pathophysiology

Background:

• Kidney size alterations are key indicators of renal pathologies.
• Dynamic parametric MRI offers a non-invasive method for kidney size measurement.
• Accurate kidney segmentation is essential for utilizing size as a biomarker.

Purpose of the Study:

• To develop and validate a deep learning model for precise kidney segmentation from dynamic parametric MRI.
• To assess the utility of automated kidney size quantification in preclinical models of renal disease.
• To establish kidney size as a sensitive biomarker for monitoring renal pathophysiology.

Main Methods:

• Utilized dynamic parametric T2 mapping MRI in rats.
• Developed a custom deep dilated U-Net (DDU-Net) architecture for kidney segmentation.
• Trained and validated the DDU-Net against manual segmentation and other models.
• Applied the DDU-Net to longitudinal in vivo MRI data with induced renal insults.

Main Results:

• The DDU-Net achieved high accuracy with a Dice coefficient of 0.98 and R-squared of 0.92.
• Automated quantification accurately detected acute kidney size changes (e.g., 11% in contrast-induced AKI).
• The model demonstrated robust performance across various interventions mimicking clinical scenarios.

Conclusions:

• Deep learning-based segmentation of dynamic parametric MRI enables accurate and automated kidney size quantification.
• This approach serves as a valuable tool for preclinical research in renal diseases.
• The methodology holds potential for developing advanced MRI-based diagnostic and monitoring tools for kidney disorders.