CNN

CNN estimates regional brain strains from head rotational velocity profiles to rapidly predict Maximum Principal Strain (MPS) and corpus callosum fiber strain for head injury assessment.

Key Features:

• Instantaneous Estimation: Conceptualizes head rotational velocity profiles as two-dimensional images to enable near-instantaneous estimation of regional brain strains.
• Strain Measures: Predicts Maximum Principal Strain (MPS) for the whole brain and corpus callosum, and fiber strain of the corpus callosum.
• High Accuracy and Reliability: Reports testing R² = 0.916 and RMSE = 0.014 for whole-brain MPS based on 2,592 training samples, and 10-fold cross-validation R² = 0.966 with RMSE = 0.013 using N = 3,069 samples.
• Independent Validation: Validated on an independent American football impact dataset comprising 314 samples.
• Pretrained CNN Models and Augmentation: Utilizes three pretrained CNN models and data augmentation across impact datasets as part of the model design and evaluation.

Scientific Applications:

• Clinical Diagnostics: Provides regional strain metrics that can inform assessment of head injuries and concussion-related biomechanical risk.
• Concussion Detection: Enables prediction of strain measures from sensor-derived rotational velocity profiles to support near-real-time concussion monitoring workflows.
• Research Transformation: Facilitates a shift in injury biomechanics research from acceleration-based metrics toward regional brain strain analysis.

Methodology:

Leveraged a dataset of N = 3,069 brain response samples to train three pretrained CNN models; explored multiple training/testing configurations using two impact datasets with data augmentation; and employed 10-fold cross-validation for performance assessment.