Advancing morphometric assessment of the aorta and left ventricle from dynamic CT: a deep learning-based study

Area of Science:

• Medical imaging
• Cardiovascular imaging
• Computational anatomy

Background:

• Manual morphometric analysis of the thoracic aorta (TA) and left ventricle (LV) is time-consuming and operator-dependent.
• Current methods are often limited to static imaging, hindering dynamic functional assessment.
• Accurate morphometrics are crucial for pre-operative planning, disease risk prediction, and device design.

Purpose of the Study:

• To develop an automated, 3D image-based framework for dynamic morphometric analysis of the TA and LV.
• To enable time-resolved quantitative assessment from ECG-gated CT datasets.
• To overcome limitations of manual, static analysis methods.

Main Methods:

• A multi-label 3D U-Net was trained for automatic segmentation of TA and LV from 50 single-phase CT scans.
• Model performance was validated on an independent multi-phase cohort of 10 patients.
• The framework generated 3D surface models for computing geometric descriptors across cardiac phases.

Main Results:

• The 3D U-Net achieved high segmentation accuracy: 97.77 ± 0.31% for TA and 91.45 ± 1.26% for LV.
• The framework successfully computed volumetric indices, displacement fields, and centerline diameters across cardiac phases in 42 patients.
• The method demonstrated robustness to variations in contrast, motion, and anatomy.

Conclusions:

• The automated framework provides a reliable and reproducible pipeline for comprehensive 3D, time-resolved morphometric analysis.
• This approach has strong potential for clinical translation in quantitative cardiac function and aortic pathophysiology assessment.
• The dynamic analysis supports improved pre-operative planning and disease risk stratification.