Abnormal functional resting-state networks in ADHD: graph theory and pattern recognition analysis of fMRI data

Area of Science:

• Neuroscience
• Computational Psychiatry
• Graph Theory Applications

Background:

• Graph theory offers tools to study the neural basis of neuropsychiatric disorders.
• Graph descriptors can potentially serve as predictive variables in classification tasks.
• Understanding brain network differences in attention-deficit/hyperactivity disorder (ADHD) is crucial.

Purpose of the Study:

• To investigate the utility of graph centrality measures in classifying children with ADHD.
• To identify brain network nodes with predictive information discriminating ADHD from typical development.
• To differentiate between ADHD subtypes (inattentive vs. combined) using network properties.

Main Methods:

• Utilized a support vector machine classifier on resting-state functional magnetic resonance imaging (fMRI) data.
• Analyzed the publicly available ADHD-200 database, a multisite dataset.
• Applied network centrality measures as predictor features to identify discriminative brain regions.

Main Results:

• Network centrality measures showed minimal predictive value for discriminating ADHD patients from healthy controls.
• Classification accuracy exceeding 65% was achieved when differentiating between inattentive and combined ADHD subtypes.
• Brain regions within motor, frontoparietal, and default mode networks were identified as containing the most predictive information.

Conclusions:

• Functional connectivity estimations are highly sensitive to sample characteristics, including acquisition protocols and clinical heterogeneity.
• Graph descriptors have limited predictive value for broad ADHD vs. typical development classification.
• Network analysis shows promise for differentiating ADHD subtypes, with specific brain networks being key.