Data Driven Models Merging Geometric, Biomechanical, and Clinical Data to Assess the Rupture of Abdominal Aortic...

Area of Science:

• Cardiovascular Surgery
• Biomedical Engineering
• Data Science

Background:

• Abdominal aortic aneurysm (AAA) rupture risk is primarily assessed by diameter, but this criterion is insufficient for preventing all ruptures.
• Rupture is a complex event influenced by multiple factors beyond simple aortic diameter.

Purpose of the Study:

• To evaluate the efficacy of machine learning (ML) models in enhancing AAA rupture risk prediction compared to diameter-based assessments.
• To compare logistic regression (LogR), linear support vector machine (SVM-Lin), non-linear support vector machine (SVM-Nlin), and Gaussian Naïve Bayes (GNB) models.

Main Methods:

• A retrospective case-control study analyzed computed tomography angiography images and clinical data from ruptured and asymptomatic AAAs.
• Finite element analysis and ML models were employed, with SHapley Additive exPlanations (SHAP) used to rank predictive factors.
• Model performance was assessed using five-fold cross-validation on the entire dataset and a subgroup with diameters ≤ 70 mm.

Main Results:

• ML models demonstrated superior accuracy in predicting AAA rupture compared to diameter alone (LogR: 90.2%, SVM-Lin: 89.5%, SVM-Nlin: 88.7%, GNB: 86.4%).
• The diameter threshold (55 mm male, 50 mm female) had 58.0% accuracy, 99.1% sensitivity, and 36.0% specificity.
• SHAP analysis identified biomechanical parameters as more relevant predictors of rupture than diameter.

Conclusions:

• A multiparameter risk assessment using ML models significantly enhances the predictive capability beyond diameter-based methods for AAA rupture.
• Further validation of this ML-based predictability method is recommended through longitudinal studies.