Atlas-Based Labeling of Resting-State fMRI

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Functional magnetic resonance imaging (fMRI) offers noninvasive brain mapping but requires manual inspection of independent components (ICs).
• Manual IC labeling is time-consuming and requires specialized expertise.
• Automating this process can enhance efficiency and reliability in neuroimaging research.

Purpose of the Study:

• To develop and validate a novel automated approach for labeling fMRI ICs.
• To establish a method based on the characteristic spatio-functional relationship of ICs.
• To reduce the time and expertise needed for fMRI data analysis.

Main Methods:

• Developed an automated approach to label fMRI ICs using their spatio-functional relationships.
• Generated a functional activation feature map based on z-score distributions across 176 subjects.
• Utilized cosine similarity to classify unlabeled ICs against the feature map.
• Validated the approach on three independent fMRI datasets (280 subjects) from the 1000 functional connectome projects.

Main Results:

• The automated approach accurately identified 9 resting-state networks and 45 ICs.
• Classification accuracy exceeded 95% across independent test datasets.
• Demonstrated a significant reduction in expert and computation time for IC labeling.
• Established an explainable relationship between functional activation and anatomically defined regions.

Conclusions:

• The proposed automated method effectively labels fMRI ICs with high accuracy and reliability.
• This approach significantly streamlines the analysis of fMRI data.
• The spatio-functional relationship provides a robust and interpretable basis for automated IC labeling.