Machine Learning and Statistical Shape Modelling Methodologies to Assess Vascular Morphology before and after Aortic...

Area of Science:

• Cardiovascular Surgery
• Medical Imaging Analysis
• Biomedical Engineering

Background:

• Statistical shape modelling (SSM) analyzes morphology and identifies shape variations.
• Hierarchical clustering identifies subgroups based on shape features.
• Aortic valve repair (AVR) involves complex patient morphology requiring detailed analysis.

Purpose of the Study:

• To evaluate aortic morphology in patients undergoing AVR using SSM and hierarchical clustering.
• To identify distinct subgroups among patients who underwent Ozaki, Ross, and valve-sparing procedures.
• To assess pre- and post-surgical morphological variability across different AVR subgroups.

Main Methods:

• Reconstruction of 3D aortic models from CT and cardiac MRI scans (n=47 pre-op, n=35 post-op).
• Application of SSM and hierarchical clustering to ascending aorta and whole aorta models.
• Comparative analysis of morphological features between different surgical subgroups.

Main Results:

• The Ross procedure subgroup exhibited unique aortic morphology, including elongated ascending aorta and wider aortic arch.
• Hierarchical clustering showed the Ross group clustering together, indicating low intra-group variability.
• Significant differences in clustering distance were observed pre-operatively (p=0.003 ascending, p=0.016 whole aorta), but not post-operatively.

Conclusions:

• SSM and hierarchical clustering are feasible for evaluating aortic morphology before and after AVR.
• This framework can aid pre-operative surgical decision-making.
• It can identify subgroups with morphology linked to poorer clinical outcomes post-operatively.