Supervised Deep Generation of High-Resolution Arterial Phase Computed Tomography Kidney Substructure Atlas

Area of Science:

• Medical Imaging
• Anatomical Atlasing
• Computational Anatomy

Background:

• The Human BioMolecular Atlas Program (HuBMAP) aims to contextualize biological findings across scales.
• Generalizing anatomical information requires robust atlas targets, but kidney imaging shows significant variability.
• This variability complicates the localization of kidney substructures for clinical analysis.

Purpose of the Study:

• To develop a high-resolution CT kidney substructure atlas template.
• To stabilize the localization of kidney substructures despite imaging variability.
• To create a generalized spatial reference for kidney anatomy.

Main Methods:

• A deep learning preprocessing technique was used to extract abdominal regions of interest.
• A deep supervised registration pipeline was employed to adapt kidney anatomical context.
• Arterial phase CT scans from 500 control subjects were registered to the atlas template.

Main Results:

• The developed atlas template successfully localized internal kidney substructures (renal cortex, medulla, pelvicalyceal system).
• Stable registration to the abdominal wall and kidney organs was achieved.
• The atlas demonstrated contextual details and generalized morphological variations across patients.

Conclusions:

• A high-resolution CT kidney substructure atlas template was successfully generated.
• The atlas provides a stabilized spatial reference for kidney anatomy, addressing imaging variability.
• This tool is applicable for generalizing morphological variations in kidney substructures across diverse patient populations.