Deep Fusion of Brain Structure-Function in Mild Cognitive Impairment

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Multimodal neuroimaging data offers complementary insights not available in single modalities.
• Deep learning models excel at uncovering complex, non-linear relationships in data.
• Existing deep models often struggle with non-Euclidean data like brain connectivity.

Purpose of the Study:

• To develop a novel deep learning framework for multimodal neuroimaging data fusion.
• To integrate neuroscience knowledge into a unified deep learning model.
• To effectively model and differentiate brain structure and function in Mild Cognitive Impairment (MCI).

Main Methods:

• Developed a graph-based deep neural network for simultaneous modeling of brain structure and function.
• Initialized graph topology using diffusion MRI (structural network).
• Iteratively updated graph incorporating functional MRI (functional network) to enhance MCI classification.

Main Results:

• Created a 'Deep Brain Connectome' by exploring deep structure-function relationships in MCI.
• The Deep Brain Connectome revealed consistent group-level alterations compared to structural networks.
• The deep model achieved 92.7% classification accuracy on the ADNI dataset for MCI detection.

Conclusions:

• The proposed graph-based deep learning approach effectively fuses multimodal neuroimaging data.
• The Deep Brain Connectome provides a powerful tool for understanding brain alterations in MCI.
• This method demonstrates significant potential for clinical diagnosis and research in neurodegenerative diseases.