Patient-specific surrogate model to predict pelvic floor dynamics during vaginal delivery

Area of Science:

• Biomedical Engineering
• Computational Mechanics
• Artificial Intelligence in Medicine

Background:

• Childbirth can cause long-term pelvic floor injuries like prolapse and incontinence.
• Computational models exist for vaginal delivery simulation but are too slow for clinical use.
• Predicting these injuries requires patient-specific biomechanical analysis.

Purpose of the Study:

• To develop an artificial intelligence (AI) pipeline for predicting pelvic floor injuries during vaginal delivery.
• To create patient-specific computational models of the pelvic floor using a mesh morphing algorithm.
• To integrate AI with finite element analysis (FEA) for rapid, accurate injury prediction.

Main Methods:

• A finite element-based machine learning approach was used.
• Thousands of childbirth simulations were performed with varied pelvic floor muscle properties.
• Machine learning models (Random Forest, XGBoost, ANNs) were trained to predict muscle stretch and nodal coordinates.

Main Results:

• The Random Forest model achieved a mean absolute error (MAE) of 0.086 mm and 11-second prediction times.
• Over 80% of nodes had prediction errors below 0.1 mm.
• The MAE for calculated muscle stretch was 0.0011.

Conclusions:

• The AI pipeline enables rapid, patient-specific prediction of pelvic floor injuries.
• This technology can assist clinicians in medical decision-making during childbirth.
• The study demonstrates the clinical feasibility of AI for predicting maternal injuries.