Physics-Based Growth and Remodeling Modeling for Virtual Abdominal Aortic Aneurysm Evolution and Growth Prediction

Area of Science:

• Biomedical Engineering
• Computational Biology
• Medical Imaging

Background:

• Computational growth and remodeling (G&R) models are crucial for studying abdominal aortic aneurysm (AAA) progression and aiding clinical decisions.
• A significant limitation in developing predictive AAA models is the scarcity of large-scale longitudinal imaging datasets.

Purpose of the Study:

• To develop a physics-based G&R framework to simulate AAA shape evolution and generate a virtual cohort.
• To address data limitations by integrating physics-based simulations with machine learning for improved AAA growth prediction.

Main Methods:

• A novel arterial G&R model was developed, incorporating elastin degradation and stress-mediated collagen production, with a modified elastin degradation formulation for realistic geometries.
• 200 distinct G&R simulations were performed, and the dataset was expanded using kriging-based spatial interpolation to create a large in silico cohort.
• Four machine learning models (DBN, RNN, LSTM, GRU) were trained and validated using the synthetic dataset combined with patient imaging data to predict maximum aneurysm diameter and growth rate.

Main Results:

• The Long Short-Term Memory (LSTM) model achieved the highest performance for maximum diameter prediction (R² = 0.92).
• The Recurrent Neural Network (RNN) model demonstrated strong overall performance, with R² = 0.90 for maximum diameter and R² = 0.89 for growth rate.
• Deep Belief Network (DBN) and Gated Recurrent Unit (GRU) models also exhibited competitive predictive capabilities.

Conclusions:

• Integrating physics-based G&R simulations with machine learning enables accurate prediction of AAA growth and maximum diameter.
• The proposed framework offers a scalable strategy for augmenting limited clinical datasets, supporting personalized risk assessment and treatment planning for AAA.