Instantaneous Whole-Brain Strain Estimation in Dynamic Head Impact.

Area of Science:

• Biomechanics
• Computational Neuroscience
• Artificial Intelligence

Background:

• Simulating head injuries is computationally intensive, limiting routine analysis.
• Accurate strain estimation is crucial for understanding traumatic brain injury (TBI).

Purpose of the Study:

• To develop a rapid and accurate method for estimating element-wise peak maximum principal strain (MPS) distribution in the brain using CNNs.
• To validate the CNN model's performance against traditional simulation methods and real-world impact data.

Main Methods:

• A convolutional neural network (CNN) was trained using 2D images derived from head impact rotational velocity and acceleration data.
• The CNN was validated using 10-fold cross-validation on a large dataset of real-world impacts (n=5661) and an independent NFL dataset (n=53).

Main Results:

• The CNN achieved over 36,000x speedup compared to traditional simulations, with >92% accuracy on real-world impacts and 96.2% on an NFL dataset.
• CNN-estimated strains showed comparable or superior concussion prediction performance to simulated strains.
• The model accurately predicted both the magnitude and distribution of peak MPS.

Conclusions:

• CNNs provide an accurate and efficient tool for estimating detailed brain strains from head impacts in contact sports.
• This AI-driven approach can accelerate TBI research and inform the design of protective gear by enabling strain-based analysis.
• The developed CNN model has the potential to transform TBI biomechanics and safety standards.