A Class Activation Map-Based Interpretable Transfer Learning Model for Automated Detection of ADHD from fMRI Data.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Attention Deficit Hyperactivity Disorder (ADHD) diagnosis can be challenging.
• Functional Magnetic Resonance Imaging (fMRI) generates complex, high-dimensional data.
• Deep Learning (DL) offers potential solutions for analyzing neuroimaging data.

Purpose of the Study:

• To develop an automated method for ADHD detection using fMRI.
• To leverage transfer learning with Convolutional Neural Networks (CNNs) for ADHD classification.
• To investigate brain regions associated with ADHD using Class Activation Maps (CAM).

Main Methods:

• Utilized a ResNet-50 pre-trained 2D-CNN model.
• Employed a transfer learning approach for ADHD classification.
• Performed 10-fold cross-validation (CV) for robust evaluation.

Main Results:

• Achieved an overall classification accuracy of 93.45% for ADHD detection.
• Class Activation Map (CAM) analysis highlighted differences in frontal, parietal, and temporal lobes.
• Demonstrated robustness against data acquisition variances and class imbalances.

Conclusions:

• Deep learning, specifically ResNet-50, is effective for automated ADHD detection from fMRI data.
• The approach addresses the curse of dimensionality inherent in neuroimaging.
• Identified specific brain areas critical for differentiating ADHD from healthy controls.