A Multichannel Deep Neural Network Model Analyzing Multiscale Functional Brain Connectome Data for Attention Deficit...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Attention deficit hyperactivity disorder (ADHD) diagnosis relies on clinical assessments, with neuroimaging offering potential biomarkers.
• Understanding brain functional connectivity across multiple scales is crucial for identifying complex neurological conditions.

Purpose of the Study:

• To develop a multichannel deep neural network (mcDNN) model for ADHD detection using multiscale brain functional connectome data.
• To evaluate the mcDNN model's performance against single-channel models and demonstrate its utility in ADHD diagnosis.

Main Methods:

• A retrospective case-control study utilized data from 973 participants in the ADHD-200 dataset.
• Multiscale functional brain connectomes were constructed, and an mcDNN model was developed integrating these connectomes with personal characteristic data (PCD).
• Model performance was assessed using cross-validation and hold-out validation, comparing mcDNN with single-channel deep neural network (scDNN) models.

Main Results:

• The mcDNN model, fusing multiscale brain connectome data and PCD, achieved the highest performance in ADHD detection during cross-validation (AUC = 0.82).
• Hold-out validation showed the mcDNN model yielding an AUC of 0.74 for ADHD detection.
• The mcDNN model significantly outperformed scDNN models that used individual scale features or PCD alone.

Conclusions:

• A novel mcDNN model was successfully developed for analyzing multiscale brain functional connectome data.
• The study demonstrated the mcDNN model's effectiveness and utility for ADHD detection.
• Fusing multiscale brain connectome data within the mcDNN framework substantially enhanced ADHD detection accuracy.