Network-based classification of ADHD patients using discriminative subnetwork selection and graph kernel PCA.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Medical Imaging Analysis

Background:

• Attention Deficit Hyperactivity Disorder (ADHD) is a common childhood disorder.
• Existing diagnostic methods struggle to identify complex brain network patterns associated with ADHD.
• ADHD involves alterations in both individual brain regions and their interconnections.

Purpose of the Study:

• To develop a novel network-based diagnostic method for ADHD.
• To overcome limitations of existing methods in detecting multi-regional brain patterns.
• To improve the accuracy and understanding of ADHD through advanced computational analysis.

Main Methods:

• A discriminative subnetwork selection method was proposed to identify frequent and distinguishing subnetworks in ADHD patients.
• Graph kernel principal component analysis (PCA) was used for feature extraction from selected subnetworks.
• Support Vector Machine (SVM) was employed for the classification of ADHD and normal control (NC) subjects.

Main Results:

• The proposed method achieved a high accuracy of 94.91% in classifying ADHD versus NC subjects on the ADHD200 dataset.
• The method successfully identified discriminative subnetworks and brain regions crucial for ADHD.
• Experimental results demonstrated significant performance improvement compared to state-of-the-art methods.

Conclusions:

• The novel network-based approach significantly enhances ADHD classification accuracy.
• The method provides valuable insights into the neurobiological underpinnings of ADHD by identifying key brain regions and connections.
• This approach offers a promising tool for more accurate ADHD diagnosis and understanding.