Recursive feature elimination for biomarker discovery in resting-state functional connectivity

Area of Science:

• Neuroimaging
• Biomarker Discovery
• Machine Learning

Background:

• Biomarker discovery in resting state functional magnetic resonance imaging (rs-fMRI) is challenging due to low signal-to-noise ratio, high dimensionality, and inter-subject variability.
• Traditional univariate analyses face limitations with multiple comparisons in complex neuroimaging datasets.

Purpose of the Study:

• To develop a data-driven machine-learning approach for identifying population differences in functional connectivity.
• To pinpoint functional connectivity features associated with symptom severity in mild traumatic brain injury (mTBI).
• To evaluate functional connections at multiple resolutions for sensitivity to recovery-related changes.

Main Methods:

• Utilized a machine-learning strategy to down-select relevant functional connectivity features from rs-fMRI data.
• Applied the method to identify altered connectivity patterns in mTBI patients.
• Compared functional connections across different scales to assess sensitivity to recovery.

Main Results:

• Identified specific functional regions with altered connectivity in mTBI, including the executive control, visual, and precuneus networks.
• Demonstrated the potential of modular network-level features for diagnostic applications.
• Found that different resolutions of functional connections vary in their sensitivity to mTBI recovery.

Conclusions:

• The proposed machine-learning approach effectively identifies functional connectivity alterations in mTBI.
• Network-level functional connectivity features show promise as diagnostic tools for predicting mTBI severity and recovery.
• Multi-resolution analysis can enhance the sensitivity of neuroimaging biomarkers for tracking patient recovery.