Learning a Phenotypic-Attribute Attentional Brain Connectivity Embedding for ADHD Classification using rs-fMRI.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Attention Deficit/Hyperactivity Disorder (ADHD) is prevalent in children.
• Functional neuroimaging, like fMRI, shows promise for ADHD diagnosis.
• Phenotypic data (age, gender) are crucial but underutilized in automated ADHD classification.

Purpose of the Study:

• To develop a novel framework integrating phenotypic information (age, gender) with brain imaging for ADHD diagnosis.
• To enhance the accuracy of automated ADHD classification using attention mechanisms and brain connectivity embeddings.

Main Methods:

• Utilized resting-state functional magnetic resonance imaging (rs-fMRI) data.
• Employed a convolutional variational autoencoder to learn brain connectivity embeddings.
• Integrated age and gender attributes via an attention mechanism, jointly optimized with the embedding learning process.

Main Results:

• Achieved a state-of-the-art average accuracy of 86.22% in classifying ADHD versus typical development control (TDC).
• Evaluation conducted across five public ADHD-200 competition datasets.
• Analysis revealed insufficient connectivity to the precuneus region in individuals with ADHD.

Conclusions:

• The proposed framework effectively integrates phenotypic data with rs-fMRI for improved ADHD diagnosis.
• The findings highlight the importance of age and gender in ADHD classification models.
• Identified precuneus connectivity as a potential biomarker for ADHD.