Mechanical Property Based Brain Age Prediction using Convolutional Neural Networks.

Area of Science:

• Neuroimaging
• Biophysics
• Machine Learning

Background:

• Brain age estimation quantifies brain structure and function relative to population norms, aiding in understanding developmental and neurodegenerative conditions.
• Magnetic Resonance Elastography (MRE) measures brain mechanical properties like stiffness and damping ratio, which change across the lifespan and reflect brain health.
• Existing methods for brain age prediction often rely on structural or functional imaging, but incorporating biomechanical data offers a novel approach.

Approach:

• Developed and trained a multi-modal 3D convolutional neural network (CNN) to model the relationship between chronological age and whole-brain mechanical properties derived from MRE.
• Evaluated the predictive performance of individual mechanical properties (stiffness, damping ratio) and brain volume, as well as multimodal combinations.
• Assessed model accuracy using Mean Absolute Error (MAE) to determine the best predictors of brain age.

Key Points:

• Stiffness maps alone predicted brain age with a Mean Absolute Error (MAE) of 3.76 years, comparable to damping ratio (MAE: 3.82) and superior to volume (MAE: 4.60).
• A multimodal approach combining stiffness and volume yielded the highest accuracy, achieving an MAE of 3.60 years.
• Including damping ratio in the multimodal model did not improve, and in some cases worsened, prediction performance.

Conclusions:

• This study demonstrates the efficacy of using brain biomechanical properties, particularly stiffness, for accurate brain age prediction.
• The developed machine learning model represents a novel advancement in predicting brain age from MRE data.
• These findings open new avenues for sensitively assessing brain integrity in individuals with neuropathologies by integrating biomechanical insights.