MAMBAxBrain: A Multi-task Neural Framework Linking Brain Functional Dynamics to Individual Fingerprints, Cognitive...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Functional magnetic resonance imaging (fMRI) data holds valuable individual, cognitive, and pathological information.
• Existing models lack a universal framework for simultaneously analyzing these diverse dimensions within fMRI data.
• There is a need for advanced computational models to integrate temporal dynamics and spatial coordination in neural activity.

Purpose of the Study:

• To introduce MAMBAxBrain, a novel multi-task neural framework designed for comprehensive fMRI data analysis.
• To jointly model the temporal dynamics and spatial coordination of neural activity using Mamba architecture and functional connectivity.
• To demonstrate the framework's efficacy across diverse fMRI-based objectives.

Main Methods:

• Integration of Mamba architecture with functional connectivity analysis within a multi-task neural framework.
• Application of the MAMBAxBrain framework to four distinct fMRI objectives: brain fingerprinting, cognitive task decoding, reaction time prediction, and schizophrenia classification.
• Evaluation of model performance against state-of-the-art methods and assessment of cross-session generalization.

Main Results:

• MAMBAxBrain achieved high accuracy across all four tested fMRI objectives, outperforming existing methods.
• The framework demonstrated robust generalization capabilities across different sessions.
• Interpretability analyses revealed distinct, biologically plausible neural circuitry engaged for each task.

Conclusions:

• The brain utilizes preferentially task-specific circuits while maintaining common representational structures across functions.
• MAMBAxBrain provides a powerful and versatile tool for multi-dimensional analysis of fMRI data.
• Findings contribute to understanding brain organization and offer insights into cognitive and pathological processes.