Brain Age Prediction Based on Resting-State Functional Connectivity Patterns Using Convolutional Neural Networks

Hongming Li, Theodore D. Satterthwaite, Yong Fan

Abstract

Brain age prediction based on neuroimaging data could help characterize both the typical brain development and neuropsychiatric disorders. Pattern recognition models built upon functional connectivity (FC) measures derived from resting state fMRI (rsfMRI) data have been successfully used to predict the brain age. However, most existing studies focus on coarse-grained FC measures between brain regions or intrinsic connectivity networks (ICNs), which may sacrifice fine-grained FC information of the rsfMRI data. Whole brain voxel-wise FC measures could provide fine-grained FC information of the brain and may improve the prediction performance. In this study, we develop a deep learning method to use convolutional neural networks (CNNs) to learn informative features from the fine-grained whole brain FC measures for the brain age prediction. Experimental results on a large dataset of resting-state fMRI demonstrate that the deep learning model with fine-grained FC measures could better predict the brain age.

Introduction

Predicting "brain age"—a biological proxy for neurodevelopmental progress—is essential for identifying atypical maturation patterns and neuropsychiatric risks. While structural MRI has traditionally been the standard for such predictions, functional connectivity (FC) derived from resting-state fMRI (rsfMRI) offers a dynamic view of brain organization. However, most current FC-based models utilize coarse-grained measures, averaging signals within predefined regions of interest (ROIs). This study argues that such spatial averaging discards critical fine-grained information. We present a deep learning framework that uses Convolutional Neural Networks (CNNs) to extract informative features from high-resolution, voxel-wise FC maps to improve brain age prediction accuracy.

Methods

The proposed methodology involves a multi-stage pipeline designed to preserve and utilize high-resolution functional data:

  1. ICN Identification: Collaborative non-negative matrix factorization (NMF) was employed to identify 56 subject-specific Intrinsic Connectivity Networks (ICNs), ensuring functional consistency across the cohort while maintaining individual anatomical nuances.
  2. Dense FC Mapping: For each of the 56 ICNs, Pearson correlation coefficients were calculated between the network's time course and every other voxel in the brain. This process generated 56 high-resolution "dense" connectivity maps per subject.
  3. CNN Architecture: These 56 maps were stacked as multi-channel inputs for a deep CNN. The architecture, shown below, comprises an initial convolutional layer followed by three Residual Blocks (ResBlocks), a fully connected layer, and a final regression output for age.
The CNN architecture utilizes 56 dense FC maps as input channels, processing them through residual blocks to predict chronological age.

Results

The model was trained and validated on the Philadelphia Neurodevelopmental Cohort (PNC), which includes rsfMRI data from 983 individuals aged 8 to 22 years. The CNN model significantly outperformed traditional linear models and graph-based architectures.

Model Features Correlation ($r$) MAE (Years)
Lasso Inter-ICN FC (Coarse) 0.417 2.45
E2Enet-sml Inter-ICN FC (Coarse) 0.361 3.02
Lasso Whole Brain FC (Fine) 0.530 2.32
CNNs (Ours) Whole Brain FC (Fine) 0.614 2.15

The Mean Absolute Error (MAE) of 2.15 years demonstrates the superior precision of the fine-grained approach. Sensitivity analysis revealed that the cingulo-opercular, default mode, and attention networks were the most influential predictors of brain age, consistent with established neurodevelopmental literature.

Examples of fine-grained voxel-wise functional connectivity maps (A, B, C) used as high-resolution inputs for the CNN model.

Discussion and Conclusion

This study highlights the importance of fine-grained spatial information in functional neuroimaging. By leveraging voxel-wise connectivity maps and hierarchical feature learning via CNNs, we achieved state-of-the-art performance in brain age prediction for an adolescent cohort. The identification of specific "clock" networks further validates the model's biological relevance. Future work will explore the integration of structural and functional data to create a more comprehensive multimodal biomarker for brain maturation.

Summary updated

Abstract

Brain age prediction based on neuroimaging data could help characterize both the typical brain development and neuropsychiatric disorders. Pattern recognition models built upon functional connectivity (FC) measures derived from resting state fMRI (rsfMRI) data have been successfully used to predict the brain age. However, most existing studies focus on coarse-grained FC measures between brain regions or intrinsic connectivity networks (ICNs), which may sacrifice fine-grained FC information of the rsfMRI data. Whole brain voxel-wise FC measures could provide fine-grained FC information of the brain and may improve the prediction performance. In this study, we develop a deep learning method to use convolutional neural networks (CNNs) to learn informative features from the fine-grained whole brain FC measures for the brain age prediction. Experimental results on a large dataset of resting-state fMRI demonstrate that the deep learning model with fine-grained FC measures could better predict the brain age.
BibTeX
@inproceedings{li2018brain,
  author = {Hongming Li and Theodore D. Satterthwaite and Yong Fan},
  title = {{Brain Age Prediction Based on Resting-State Functional Connectivity Patterns Using Convolutional Neural Networks}},
  booktitle = {ISBI},
  year = {2018},
  url = {https://openreview.net/forum?id=ym4D0HFja2Q},
}
Figure