Convolutional neural network for efficient estimation of regional brain strains.

Area of Science:

• Biomechanics
• Computational Neuroscience
• Machine Learning

Background:

• Traditional head injury models for concussion biomechanics are too slow for real-world applications.
• Accurate and rapid estimation of regional brain strains is crucial for understanding head injury mechanisms.

Purpose of the Study:

• To develop a fast and accurate method for estimating regional brain strains using a convolutional neural network (CNN).
• To conceptualize head rotational velocity profiles as 2D images for CNN input, enabling instant strain prediction.

Main Methods:

• A CNN was developed and trained using two impact datasets with data augmentation.
• The CNN was evaluated on its ability to predict maximum principal strain (MPS) of the whole brain, MPS of the corpus callosum, and fiber strain of the corpus callosum.
• The model was further validated using an independent dataset of 314 American football impacts.

Main Results:

• The CNN achieved a testing R-squared of 0.916 and RMSE of 0.014 for whole brain MPS using 2592 training samples.
• With all available data (3069 samples), the CNN achieved an R-squared of 0.966 and RMSE of 0.013 in a 10-fold cross-validation.
• The model demonstrated high accuracy in predicting various regional brain strain measures.

Conclusions:

• The developed CNN technique offers instant and accurate estimation of regional brain strains, overcoming the speed limitations of traditional models.
• This approach has the potential to enable clinical diagnostic capabilities for head injuries and advance concussion detection.
• It may shift the focus of injury studies from acceleration-based metrics to regional brain strains, providing deeper insights into head trauma.