Three-dimensional convolutional neural network-based classification of chronic kidney disease severity using kidney...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Nephrology

Background:

• Chronic kidney disease (CKD) affects millions globally, necessitating accurate and efficient diagnostic tools.
• Current CKD diagnosis relies on estimated glomerular filtration rate (eGFR) and imaging, but advanced AI methods offer potential improvements.
• Magnetic Resonance Imaging (MRI) provides detailed anatomical and functional information crucial for assessing kidney health.

Purpose of the Study:

• To develop and evaluate a 3D convolutional neural network (CNN) model for classifying CKD severity.
• To assess the model's performance using various MRI sequences (T1-weighted in-phase, opposed-phase, water-only) and kidney laterality (unilateral vs. bilateral).
• To determine the optimal configuration of the AI model for accurate CKD staging.

Main Methods:

• A 3D CNN model was trained and tested on MRI data from 322 patients with varying CKD severities (mild, moderate, severe).
• The model utilized Dixon-based T1-weighted in-phase (IP), opposed-phase (OP), and water-only (WO) imaging sequences.
• Performance was evaluated using both unilateral and bilateral kidney inputs, and across different imaging sequences.

Main Results:

• The best model performance was achieved using bilateral kidneys and in-phase (IP) MRI images, yielding an accuracy of 0.862 ± 0.036.
• Bilateral kidney models demonstrated superior accuracy compared to unilateral kidney models.
• The deep learning approach showed promising results in classifying CKD severity based on MRI findings.

Conclusions:

• A 3D CNN model applied to kidney MRI is a viable tool for classifying CKD severity.
• Utilizing bilateral kidney information and specific MRI sequences (IP) enhances classification accuracy.
• This AI-driven method offers a potential non-invasive approach for objective CKD staging.