Multitask fMRI Data Classification via Group-Wise Hybrid Temporal and Spatial Sparse Representations.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• Task-based functional magnetic resonance imaging (tfMRI) is crucial for studying brain activity during cognitive tasks.
• Existing deep learning models for fMRI analysis require large datasets and often lack explainability.
• Challenges in fMRI data include high dimensionality, low signal-to-noise ratio, and interindividual variability.

Purpose of the Study:

• To develop a computational framework for classifying tfMRI data based on signal composition patterns.
• To identify key functional components that differentiate various task states.
• To address challenges of small sample sizes and explainability in fMRI analysis.

Main Methods:

• Proposed a group-wise hybrid temporal and spatial sparse representations (HTSSR) framework.
• Utilized Human Connectome Project (HCP) tfMRI data for validation.
• Applied the framework to identify and differentiate multitask fMRI signal patterns.

Main Results:

• Achieved an average classification accuracy of 96.67% for multitask fMRI data.
• Successfully identified key components for differentiating tasks, demonstrating method explainability.
• Reliably detected both task-related components and resting-state networks (RSNs).

Conclusions:

• The proposed HTSSR framework effectively classifies tfMRI data and provides interpretable results.
• The method is suitable for analyzing neuroimaging datasets with small sample sizes.
• Potential applications include fMRI data quality control and biomarker discovery for brain disorders.