Template independent component analysis with spatial priors for accurate subject-level brain network estimation and...

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Statistical Modeling

Background:

• Independent Component Analysis (ICA) is widely used for analyzing functional magnetic resonance imaging (fMRI) data to identify brain networks.
• Standard ICA provides reliable group-level results but often yields noisy single-subject estimates.
• Existing hierarchical models like Template ICA improve subject-level analysis but unrealistically assume spatial independence of subject effects.

Purpose of the Study:

• To introduce spatial template ICA (stICA), a novel hierarchical ICA model.
• To enhance estimation efficiency by incorporating spatial priors into the Template ICA framework.
• To improve the identification of brain regions engaged in functional networks using an excursions set approach on the joint posterior distribution.

Main Methods:

• Development of the spatial template ICA (stICA) model, integrating spatial priors.
• Derivation of an efficient expectation-maximization algorithm for parameter estimation and posterior moment calculation.
• Application of an excursions set approach for identifying engaged brain regions, leveraging spatial dependencies and avoiding massive multiple comparisons.

Main Results:

• stICA demonstrated more accurate and reliable estimates compared to benchmark approaches on both simulated and Human Connectome Project fMRI data.
• The method successfully identified larger and more dependable areas of brain network engagement.
• The derived expectation-maximization algorithm proved computationally tractable, converging within 12 hours for whole-cortex fMRI data.

Conclusions:

• stICA offers a powerful and efficient approach for subject-level functional brain network analysis in fMRI.
• Incorporating spatial dependencies significantly improves the reliability and accuracy of network estimation.
• The computational efficiency makes stICA a practical tool for large-scale neuroimaging studies.