Multimodal Brain Age Prediction with Feature Selection and Comparison.

Area of Science:

• Neuroscience
• Radiology
• Biomarkers

Background:

• Brain age, estimated from brain imaging, and its gap from chronological age, are potential biomarkers.
• These biomarkers can indicate typical development, abnormal aging, and neuropsychiatric issues.

Purpose of the Study:

• To predict brain age using multimodal brain imaging data.
• To compare the performance of individual and combined data modalities.
• To evaluate feature selection and machine learning models for brain age prediction.

Main Methods:

• Utilized multimodal brain imaging data (gray matter density, anatomical features, functional connectivity) from 1417 participants (ages 8-22).
• Compared individual and combined data modalities using linear support vector and partial least squares regression.
• Applied feature selection to identify optimal predictors for brain age.

Main Results:

• Feature selection and multimodal data improved brain age prediction accuracy.
• Achieved a mean absolute error of 1.22 years using 188 selected features, outperforming individual data sources.
• The bias-corrected brain age gap correlated significantly with attention and motor speed.

Conclusions:

• Multimodal imaging with feature selection enhances brain age prediction.
• Traditional machine learning models perform comparably to deep learning for this sample size.
• The brain age gap serves as a valuable indicator of cognitive and motor functions.