Machine Learning Prediction of Vaginal Tissue Tears Using Finite Element Simulations Informed by Planar Biaxial...

Area of Science:

• Biomechanical engineering
• Computational modeling
• Medical device simulation

Background:

• Childbirth-related vaginal tearing is common, with significant long-term implications.
• Understanding the mechanical factors driving tear progression is limited by in vivo study difficulties.
• Finite element (FE) models and machine learning (ML) offer potential solutions for studying these complex biomechanical behaviors.

Purpose of the Study:

• To investigate how tear geometry and fiber orientation affect local stress and strain patterns in vaginal tissue.
• To determine if ML models can accurately predict tissue responses from FE simulation data.
• To develop a computational tool for evaluating childbirth injury risk.

Main Methods:

• A Holzapfel-Gasser-Ogden (HGO) hyperelastic constitutive model was calibrated using swine vaginal tissue data.
• FE simulations were performed for various tear orientations, sizes, and fiber alignments.
• Four ML models (including XGBoost) were trained and validated on the FE-generated dataset.

Main Results:

• FE simulations accurately reproduced experimental results, validating the HGO model.
• Extreme Gradient Boosting (XGBoost) demonstrated the highest predictive accuracy for mechanical and geometric outputs.
• Tear orientation was identified as the most significant factor influencing local stress and strain at the tear boundary.

Conclusions:

• An integrated FE-ML approach provides rapid and accurate predictions of vaginal tissue behavior under diverse tear conditions.
• This method bypasses the need for additional simulations or experiments, utilizing synthetic data.
• The approach serves as a valuable tool for injury risk assessment and clinical decision-making in maternal health.