3D Adversarial Segmentation of Kidney-Transplant Across Multiple MRI Sequences Using Probabilistic and Anatomical...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Nephrology

Background:

• Accurate kidney segmentation in kidney-transplant patients is vital for quantitative graft assessment.
• Challenges include low tissue contrast, intensity inhomogeneity, and anatomical variability in MRI.
• Existing methods struggle with precise delineation in complex transplant cases.

Purpose of the Study:

• To develop and evaluate a robust 3D adversarial segmentation framework for kidney transplants.
• To integrate probabilistic appearance and anatomical shape priors into a residual conditional generative adversarial network (GAN).
• To enhance boundary delineation under challenging MRI conditions.

Main Methods:

• A 3D adversarial segmentation framework using a residual conditional GAN.
• Incorporation of probabilistic appearance and anatomical shape priors.
• Evaluation on 100 kidney-transplant patients across T2-weighted, BOLD-MRI, and DW-MRI sequences.

Main Results:

• Achieved high mean Dice scores: 90.86% (T2w), 92.02% (BOLD-MRI), 94.00% (DW-MRI).
• Demonstrated robust and consistent performance across heterogeneous MRI acquisitions.
• Prior guidance improved segmentation stability and anatomical consistency, especially in low-contrast scans.

Conclusions:

• The proposed framework enables reliable kidney delineation across multiple MRI sequences.
• Supports consistent extraction of quantitative imaging biomarkers for noninvasive graft assessment.
• Facilitates longitudinal monitoring of renal transplant patients.