Brain Age Prediction With Morphological Features Using Deep Neural Networks: Results From Predictive Analytic...

Area of Science:

• Neuroimaging
• Machine Learning
• Computational Neuroscience

Background:

• Structural magnetic resonance imaging (MRI) and machine learning (ML) algorithms are key to studying brain morphological changes across the lifespan.
• Open access imaging datasets and international collaborations have accelerated brain structure characterization and brain age prediction.
• Brain age prediction serves as a valuable metric for assessing neurodevelopment and aging.

Purpose of the Study:

• To present a predictive model based on deep neural networks (DNNs) for brain age prediction.
• To evaluate the performance of the proposed DNN architecture against other common ML algorithms.
• To develop a framework for robust statistical evaluation of feature importance for clinical interpretability.

Main Methods:

• Utilized structural MRI scans from 2638 healthy individuals across 17 collection sites.
• Extracted morphological descriptors using FreeSurfer software.
• Compared a proposed DNN architecture with Random Forest (RF), Support Vector Regression (SVR), and Lasso algorithms.

Main Results:

• The DNN models achieved the highest performance, with a Mean Absolute Error (MAE) of 4.6 on the hold-out test set.
• DNNs significantly outperformed RF, SVR, and Lasso in brain age prediction accuracy.
• A comprehensive ML framework was established for feature importance analysis.

Conclusions:

• Deep neural networks demonstrate superior efficacy for accurate brain age prediction compared to traditional ML methods.
• The developed ML framework facilitates clinical interpretability by evaluating feature importance.
• This study advances the field of neuroimaging analysis for understanding brain aging and development.