An integrated finite element method and machine learning algorithm for brain morphology prediction.

Area of Science:

• Computational neuroscience
• Developmental biology
• Biophysics

Background:

• Human brain development involves intricate cortical folding from a smooth to a convoluted surface.
• Computational modeling aids understanding of cortical folding but faces challenges in simulation scale and cost.
• Existing models struggle to provide reliable predictions for brain folding with limited computational resources.

Purpose of the Study:

• To develop a machine-learning-based finite element surrogate model for expedited brain development simulations.
• To predict brain folding morphology and explore underlying developmental mechanisms.
• To overcome computational limitations in simulating massive brain development.

Main Methods:

• Generated massive finite element method (FEM) simulations of brain development using growth models with adjustable surface curvature.
• Trained and validated a Generative Adversarial Network (GAN)-based machine learning model using FEM-generated data.
• Utilized machine learning for data augmentation and prediction to create a surrogate model.

Main Results:

• The machine learning models accurately predicted complex brain folding patterns, including 3-hinge gyral folds.
• FEM simulations provided the computational data for training and validating the machine learning models.
• The study demonstrated the feasibility of using machine learning to predict brain folding morphology.

Conclusions:

• The developed machine-learning surrogate model significantly expedites brain computational simulations.
• The approach offers a promising method for predicting brain development based on fetal configurations.
• This study validates the use of machine learning in understanding the mechanisms of cortical folding.