Deep Spatio-Temporal Representation and Ensemble Classification for Attention Deficit/Hyperactivity Disorder

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Attention Deficit/Hyperactivity Disorder (ADHD) is a complex neurodevelopmental disorder with heterogeneous characteristics.
• Current ADHD diagnosis relies on extensive clinical data analysis, including behavioral and psychological assessments.
• Functional Magnetic Resonance Imaging (fMRI) offers a non-invasive approach for studying brain cognition.

Purpose of the Study:

• To develop an advanced algorithm for improving ADHD classification accuracy using fMRI data.
• To leverage deep learning and ensemble methods for robust feature extraction and classification.

Main Methods:

• A Convolutional Denoising Autoencoder (CDAE) was developed to extract spatial features from fMRI data.
• Adaptive Boosting Decision Trees (AdaDT) were employed to classify the features extracted by the CDAE.
• The proposed algorithm was validated on the ADHD-200 test dataset.

Main Results:

• The CDAE-AdaDT algorithm demonstrated improved ADHD classification accuracy compared to existing state-of-the-art methods.
• The method achieved higher average accuracy across individual sites and overall.
• The algorithm maintained a balance between classification specificity and sensitivity.

Conclusions:

• The proposed CDAE-AdaDT algorithm represents a significant advancement in ADHD classification using fMRI.
• This approach offers a promising tool for more accurate and reliable ADHD diagnosis.
• The findings highlight the potential of integrating deep learning with ensemble methods in neuroimaging research.