Multi-contrast computed tomography healthy kidney atlas.

Area of Science:

• Medical Imaging
• Anatomical Atlasing
• Computational Anatomy

Background:

• Integrating multi-scale biological data requires robust anatomical frameworks.
• Existing abdominal CT atlases lack multi-contrast support and detailed kidney morphometry.
• Normal kidney anatomy exhibits significant variation influenced by body size, sex, and imaging parameters.

Purpose of the Study:

• To develop a high-resolution CT retroperitoneal atlas optimized for kidneys across multiple contrast phases.
• To create a volumetric atlas framework for integrating and visualizing normal kidney anatomical variations.
• To establish a deep learning-based pipeline for automated organ extraction and registration within the atlas framework.

Main Methods:

• A deep learning-based volume interest extraction method was employed for slice localization and cropping.
• An automated two-stage hierarchical registration pipeline (DEEDS affine and non-rigid) was used for volume registration.
• Multi-contrast CT scans from 500 subjects were processed to generate and evaluate the atlas framework.

Main Results:

• The PDD-Net with affine registration showed the best overall mean DICE score (0.540 ± 0.275) for multi-organ label transfer in the portal venous phase.
• The pipeline achieved a median DICE score over 0.8 for transferring kidney information to the atlas space.
• DEEDS registration demonstrated stable performance across all contrast phases, accurately mapping kidney boundaries, while PDD-Net showed stability in arterial and venous phases.

Conclusions:

• The developed CT atlas framework successfully integrates normal kidney variations across diverse demographics and contrast modalities.
• The atlas provides a valuable tool for understanding population-specific anatomical variability in kidneys.
• The automated registration pipeline ensures stable and generalizable mapping of kidney data within the atlas space.