MADE-for-ASD: A multi-atlas deep ensemble network for diagnosing Autism Spectrum Disorder.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Autism Spectrum Disorder (ASD) diagnosis is complex and time-consuming.
• Existing methods lack efficiency and scalability for early detection.
• Automated diagnostic tools are needed to bridge this gap.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for efficient and accurate ASD diagnosis.
• To integrate multi-atlas functional magnetic resonance imaging (fMRI) data with demographic information.
• To improve upon existing ASD diagnostic accuracy using automated methods.

Main Methods:

• Proposed a multi-atlas deep ensemble network (MADE-for-ASD) for ASD diagnosis.
• Integrated resting-state fMRI data from the ABIDE I dataset with demographic information.
• Utilized a weighted deep ensemble network to combine multiple brain atlases.

Main Results:

• Achieved 75.20% accuracy on the entire ABIDE I dataset and 96.40% on a specific subset.
• Outperformed prior ASD diagnosis accuracy in fMRI studies by 4.4 percentage points.
• Demonstrated high sensitivity (82.90%) and specificity (69.70%) on the full dataset.

Conclusions:

• The MADE-for-ASD system offers a cost-effective, efficient, and scalable approach to ASD diagnosis.
• The model's integration of fMRI and demographic data provides a holistic patient profile.
• This automated system has the potential to significantly improve early ASD detection and intervention.