Dynamic functional connectivity analysis with temporal convolutional network for attention deficit/hyperactivity...

Area of Science:

• Neuroimaging
• Machine Learning
• Psychiatry

Background:

• Dynamic functional connectivity (dFC) in resting-state fMRI (rs-fMRI) offers insights into brain activity abnormalities over time.
• Existing deep learning methods for dFC analysis in Attention Deficit/Hyperactivity Disorder (ADHD) often overlook long-range temporal dependencies.

Purpose of the Study:

• To propose a novel Temporal Dependence neural Network (TDNet) for enhanced FC representation learning and temporal-dependence tracking in rs-fMRI data.
• To achieve automated and accurate identification of ADHD by analyzing dynamic brain activity patterns.

Main Methods:

• rs-fMRI time series are segmented, and an FC generation module creates discriminative dynamic FCs for each segment.
• A Temporal Convolutional Network (TCN) with dilated convolutions is employed to capture long-range temporal dependencies.
• Fully connected layers are utilized for the final disease prediction.

Main Results:

• Incorporating dynamic characteristics of rs-fMRI data significantly improves diagnostic performance.
• Data-driven dynamic FC networks are more informative than traditional Pearson correlation-based methods.
• The proposed TDNet model demonstrates superior performance in ADHD identification compared to state-of-the-art methods on the ADHD-200 database.

Conclusions:

• Dynamic functional connectivity analysis using TDNet provides a more effective approach for ADHD identification.
• The TDNet model's ability to capture long-range temporal patterns is crucial for understanding ADHD-related brain activity abnormalities.
• This data-driven approach offers a promising tool for the automated diagnosis of ADHD.