Autism spectrum disorder diagnosis based on deep unrolling-based spatial constraint representation.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Biomedical Engineering

Background:

• Accurate autism spectrum disorder (ASD) diagnosis is vital for treatment and prognosis.
• Functional brain networks (FBNs) from fMRI are used for ASD diagnosis, but existing methods struggle with non-linear relationships and separate network construction/classification steps.
• This separation leads to inter-subject variability and reduced statistical power in FBNs.

Purpose of the Study:

• To propose a novel end-to-end framework for ASD recognition by integrating a new FBNs construction model with a convolutional classifier.
• To address limitations of existing methods in capturing non-linear relationships and reducing inter-subject variability in FBNs.

Main Methods:

• Developed the deep unrolling-based spatial constraint representation (DUSCR) model for FBNs construction, utilizing a proximal gradient descent algorithm and deep unrolling.
• Integrated DUSCR with a convolutional prototype learning classifier for an end-to-end ASD recognition framework.
• Preprocessed resting-state fMRI data into time series and 3D coordinates for input into the DUSCR model.

Main Results:

• The proposed DUSCR model integrated with a convolutional classifier demonstrated significant improvements in model performance and classification accuracy on the ABIDE I dataset.
• The end-to-end framework effectively captures non-linear relationships within fMRI data.
• Reduced inter-subject variability in FBNs estimation was observed, enhancing statistical power.

Conclusions:

• The proposed deep unrolling-based spatial constraint representation (DUSCR) model offers a more effective approach to constructing functional brain networks for ASD diagnosis.
• The integrated end-to-end framework significantly enhances ASD recognition accuracy compared to traditional methods.
• This approach holds promise for improving diagnostic tools for autism spectrum disorder.