A predictive surrogate model based on linear and nonlinear solution manifold reduction in cardiovascular FSI: A...

Area of Science:

• Biomedical Engineering
• Computational Fluid Dynamics
• Cardiovascular Mechanics

Background:

• Abdominal aortic aneurysms (AAAs) cause significant hemodynamic alterations.
• Accurate AAA simulation requires patient-specific models due to geometric complexity.
• Understanding fluid-structure interaction (FSI) is crucial for predicting AAA progression.

Purpose of the Study:

• To enhance predictive capabilities for flow and structural indices in AAA FSI simulations.
• To compare the efficacy of two surrogate models: Proper Orthogonal Decomposition + Long Short-Term Memory (POD + LSTM) and Convolutional Neural Network + Long Short-Term Memory (CNN + LSTM).
• To analyze performance under varying physiological conditions (rest and exercise).

Main Methods:

• Development and implementation of POD + LSTM and CNN + LSTM surrogate models.
• Fluid-structure interaction (FSI) simulations using patient-specific abdominal aorta models.
• Comparative analysis of model accuracy, computational cost, and predictive power.

Main Results:

• Both POD + LSTM and CNN + LSTM models demonstrated potential for accurate FSI simulation.
• Performance metrics indicated varying strengths and limitations for each surrogate model.
• The study provides insights into the suitability of each model for personalized cardiovascular simulations.

Conclusions:

• Surrogate models like POD + LSTM and CNN + LSTM offer promising avenues for efficient patient-specific cardiovascular simulations.
• Accurate prediction of hemodynamic alterations in AAAs is feasible with advanced computational approaches.
• Further research is needed to optimize these models for clinical application.