Machine Learning-Based Segmentation of the Thoracic Aorta with Congenital Valve Disease Using MRI.

Area of Science:

• Cardiovascular Imaging
• Biomedical Engineering
• Computational Fluid Dynamics

Background:

• Bicuspid aortic valve (BAV) subjects require regular imaging for dysfunction surveillance.
• Manual aortic segmentation for hemodynamic assessment is time-consuming and prone to bias.
• Understanding BAV morphology's impact on aortic shear stress is crucial for predicting valve incompetence.

Purpose of the Study:

• To employ machine learning (ML)-based segmentation for improved BAV assessment.
• To quantify the relationship between BAV morphology and vortical structures.
• To analyze how BAV morphology influences aortic shear stress and susceptibility to incompetence.

Main Methods:

• ML segmentation model trained on whole-body CT data.
• Acquired MRI from subjects with tricuspid aortic valves (TAV) and BAV.
• Utilized 4D-PCMRI for quantifying vortical structures and wall shear stress.

Main Results:

• ML model achieved a high Dice score (0.86) for heart segmentation; thoracic aorta segmentation was poorer (0.72).
• Tricuspid aortic valves (TAVs) showed symmetric wall shear stress.
• Bicuspid aortic valves (BAVs) exhibited asymmetric wall shear stress, with elevated tangential shear stress opposite fused leaflets due to helical flow.

Conclusions:

• ML-based segmentation reduces assessment runtime and identifies the significance of tangential wall shear stress in BAVs.
• Asymmetric shear stress in BAVs is linked to valve incompetence progression.
• Findings may guide surgical interventions for BAV patients.