Multimodal Sparse Classifier for Adolescent Brain Age Prediction.

Area of Science:

• Neuroscience
• Machine Learning
• Developmental Psychology

Background:

• Understanding healthy brain development is crucial for mapping transformations and connectivity from childhood to adulthood.
• Functional magnetic resonance imaging (fMRI) provides valuable data for studying brain activity and connectivity.

Purpose of the Study:

• To develop a sparse machine learning solution for analyzing whole-brain functional connectivity in adolescents.
• To predict brain age using multimodal fMRI data from the Philadelphia Neurodevelopmental Cohort (PNC).

Main Methods:

• Utilized resting-state fMRI (rs-fMRI), working memory n-back task fMRI (nb-fMRI), and emotion identification task fMRI (em-fMRI) data.
• Developed a sparse Extreme Learning Machine (ELM) classifier based on residual errors to handle high-dimensional fMRI data with many irrelevant and correlated features.
• The proposed method prunes redundant features and output weights to overcome overlearning.

Main Results:

• The multimodal sparse ELM classifier demonstrated high classification accuracy in predicting brain age.
• The method effectively extracted relevant features from complex, high-dimensional fMRI datasets.
• The sparse ELM approach outperformed conventional ELM and sparse Bayesian learning ELM in classification tasks.

Conclusions:

• The proposed sparse ELM method is a competitive and effective approach for analyzing adolescent brain connectivity and predicting brain age.
• This technique addresses the challenges of high-dimensional fMRI data, offering improved feature extraction and classification accuracy.
• The findings contribute to a better understanding of brain development and connectivity patterns during adolescence.