Application of Convolutional Recurrent Neural Network for Individual Recognition Based on Resting State fMRI Data

Area of Science:

• Neuroimaging
• Machine Learning
• Neuroscience

Background:

• Individual variability in brain function is crucial but often overlooked in fMRI studies.
• Previous work used static functional connectomes and Recurrent Neural Networks (RNNs) for individual identification.
• Temporal dynamics in brain activity are increasingly recognized as important for understanding individual differences.

Purpose of the Study:

• To apply Convolutional RNN (ConvRNN) for individual identification using resting-state fMRI data.
• To evaluate ConvRNN's performance compared to conventional RNN in capturing individual brain patterns.
• To explore the visualization capabilities of ConvRNN for fMRI data analysis.

Main Methods:

• Utilized resting-state functional Magnetic Resonance Imaging (fMRI) data.
• Applied Convolutional Recurrent Neural Networks (ConvRNN) for individual identification.
• Compared ConvRNN performance against conventional RNN models.

Main Results:

• ConvRNN achieved higher identification accuracy than conventional RNN.
• The improvement is attributed to ConvRNN's enhanced extraction of local spatial and temporal features between brain regions (ROIs).
• ConvRNN outputs allowed for visualization of informative spatial and temporal patterns.

Conclusions:

• ConvRNN is a promising method for individual identification from resting-state fMRI data.
• Integrating spatial and temporal information via ConvRNN enhances the analysis of individual brain variability.
• The visualization capabilities of ConvRNN offer new avenues for fMRI data interpretation.