Explainable Multimodal Deep Dictionary Learning to Capture Developmental Differences From Three fMRI Paradigms.

Area of Science:

• Neuroscience
• Developmental Neuroscience
• Machine Learning in Neuroscience

Background:

• Multimodal approaches integrate diverse data for comprehensive neuroscience insights.
• Limited research exists on multimodal analysis of brain developmental changes.

Purpose of the Study:

• To develop an explainable multimodal deep dictionary learning method.
• To uncover commonalities and specificities across different brain data modalities.
• To identify brain developmental differences using multimodal fMRI data.

Main Methods:

• Proposed an explainable multimodal deep dictionary learning framework.
• Utilized a sparse deep autoencoder for data encoding.
• Trained a shared dictionary and modality-specific sparse representations on fMRI data from three paradigms (two tasks, resting state).

Main Results:

• The model achieved superior reconstruction performance.
• Identified age-related differences in recurring brain activity patterns.
• Found that children and young adults exhibit distinct functional connectivity patterns during tasks and rest, with children showing more diffuse and young adults more focused patterns.

Conclusions:

• The multimodal deep learning method effectively reveals developmental differences in brain networks.
• Understanding these age-related network changes is crucial for comprehending neural circuit formation and development.
• The findings highlight differences in functional connectivity strategies between children and young adults.