Moving Beyond Functional Connectivity: Time-Series Modeling for fMRI-Based Brain Disorder Classification.

Area of Science:

• Neuroimaging
• Machine Learning
• Computational Neuroscience

Background:

• Functional magnetic resonance imaging (fMRI) uses blood-oxygen-level-dependent (BOLD) signals for non-invasive brain disorder classification.
• Current methods often rely on functional connectivity (FC) via Pearson correlation, which simplifies 4D BOLD data into static 2D matrices, losing temporal dynamics and linear relationships.

Purpose of the Study:

• To benchmark state-of-the-art temporal models against traditional FC-based approaches for fMRI brain disorder classification.
• To introduce DeCI, a novel framework that integrates cycle-drift decomposition and channel-independence for enhanced fMRI analysis.

Main Methods:

• Benchmarking of advanced time-series models (PatchTST, TimesNet, TimeMixer) on raw BOLD signals across five public fMRI datasets.
• Development and application of the DeCI framework, featuring Cycle and Drift Decomposition for each Region of Interest (ROI) and Channel-Independence modeling.
• Comparative analysis against traditional FC-based methods and other temporal baselines.

Main Results:

• State-of-the-art temporal models consistently outperformed traditional FC-based approaches in classification accuracy.
• The proposed DeCI framework demonstrated superior classification accuracy and generalization capabilities compared to both FC and temporal baselines.
• Directly modeling temporal information, including oscillatory fluctuations and slow baseline trends, proved crucial for improved performance.

Conclusions:

• End-to-end temporal modeling of BOLD signals offers a more effective approach for fMRI-based brain disorder classification than static FC methods.
• The DeCI framework provides a robust and accurate method for capturing complex brain dynamics, advancing the field of neuroimaging analysis.
• Findings advocate for a paradigm shift towards temporal modeling in fMRI analysis to better understand brain function and dysfunction.