Encoding and Decoding of Brain Dynamic Functional Connectivity for ADHD Diagnosis

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• Dynamic functional connectivity (FC) changes correlate with cognitive functions.
• Traditional sliding window techniques for dynamic FC face challenges like distributional shifts and high dimensionality.
• Accurate modeling of dynamic FC is crucial for understanding brain function and diagnosing neurological disorders.

Purpose of the Study:

• To introduce BRAINMAP (Bi-level Representation using Attention for INterpretability with Mamba-Aided Prediction), a novel method for modeling dynamic brain functional connectivity.
• To address the limitations of the sliding window technique, specifically distributional shifts and high dimensionality.
• To improve the accuracy of Attention Deficit Hyperactivity Disorder (ADHD) detection using dynamic FC.

Main Methods:

• BRAINMAP utilizes Optimal Transport to correct distributional shifts across sliding windows.
• It employs Graph Neural Networks (GNNs) combined with an attention mechanism and Mamba block to capture spatiotemporal features from functional MR images.
• A Top-K sliding window feature selection algorithm is introduced to induce sparsity in dynamic FC.

Main Results:

• BRAINMAP demonstrated superior performance in ADHD detection compared to existing state-of-the-art dynamic FC models, with accuracy improvements of 3% to 12% across three datasets (ADHD-200, UCLA, CNI-TLC).
• The model identified robust biomarkers, particularly the connection between the dorsal attention network and the visual network.
• An association study confirmed the clinical relevance of the identified biomarkers.

Conclusions:

• BRAINMAP offers a significant advancement in modeling dynamic functional connectivity for brain imaging analysis.
• The proposed method effectively addresses key challenges in dynamic FC analysis, leading to improved diagnostic accuracy for ADHD.
• The identified biomarkers, such as the dorsal attention-visual network connection, hold clinical relevance for ADHD diagnosis and understanding its underlying neural mechanisms.