CATD: Unified Representation Learning for EEG-to-fMRI Cross-Modal Generation.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Biomedical Engineering

Background:

• Multi-modal neuroimaging is vital for understanding brain function and pathology but faces challenges due to high costs and limited availability of certain techniques.
• Integrating diverse imaging modalities like fMRI and EEG offers a more comprehensive view, overcoming individual limitations.

Purpose of the Study:

• To develop a novel framework for cross-modal synthesis of neuroimaging data, specifically generating fMRI-BOLD signals from more accessible EEG signals.
• To enhance the temporal resolution of neuroimaging analysis for capturing dynamic brain activity.

Main Methods:

• Proposed the Condition-Aligned Temporal Diffusion (CATD) framework for end-to-end cross-modal synthesis.
• Introduced the Conditionally Aligned Block (CAB) to align heterogeneous neuroimages into a unified potential space.
• Integrated the Dynamic Time-Frequency Segmentation (DTFS) module to leverage EEG for improving BOLD signal temporal resolution.

Main Results:

• Achieved a 9.13% improvement in brain activity state prediction accuracy (reaching 69.8%).
• Enhanced diagnostic accuracy for brain disorders by 4.10% (reaching 99.55%).
• Successfully identified abnormal brain regions and improved the temporal resolution of BOLD signals.

Conclusions:

• The CATD framework establishes a new paradigm for cross-modal neuroimaging synthesis by unifying data into a potential representation space.
• Demonstrated significant potential for medical applications, including improved prediction of diseases like Parkinson's and identification of abnormal brain regions.