Extreme learning machine-based classification of ADHD using brain structural MRI data.

Area of Science:

• Neuroimaging
• Machine Learning
• Psychiatry

Background:

• Attention-deficit/hyperactivity disorder (ADHD) diagnosis is a significant clinical challenge.
• Structural MRI data reveals cortical features associated with ADHD.
• Existing diagnostic algorithms suffer from inefficiencies like long training times and complex parameter selection.

Purpose of the Study:

• To introduce an Extreme Learning Machine (ELM) model for automated, efficient, and objective ADHD diagnosis.
• To evaluate the computational efficiency and sample size robustness of ELM versus Support Vector Machine (SVM).
• To identify specific brain segments implicated in ADHD.

Main Methods:

• Acquired high-resolution 3D MRI data from 55 ADHD subjects and 55 healthy controls.
• Calculated 340 cortical features from 68 brain segments using automated FreeSurfer analysis.
• Employed F-score and Sequential Floating Search (SFS) for optimal feature selection, with classification accuracy assessed via leave-one-out cross-validation for both ELM and SVM.

Main Results:

• Achieved 90.18% accuracy for ELM, outperforming SVM-Linear (84.73%) and SVM-RBF (86.55%).
• ELM demonstrated superior computational efficiency and robustness to sample size variations compared to SVM.
• Significant differences between ADHD and control groups were most prominent in the frontal, temporal, occipital lobes, and insula.

Conclusions:

• The ELM-based algorithm significantly outperforms traditional SVM for ADHD diagnosis.
• ELM presents a promising tool for clinical ADHD diagnosis and broader neurological disease investigation.