Sparse Multi-view Task-Centralized Learning for ASD Diagnosis.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Early diagnosis of autism spectrum disorder (ASD) remains a significant challenge.
• Neuroimaging data offers potential for identifying biomarkers, but effective analysis methods are needed.
• ASD's known age- and sex-related variations complicate diagnostic approaches.

Purpose of the Study:

• To propose a novel Sparse-MVTC classification method for computer-assisted diagnosis of ASD.
• To leverage multi-view functional magnetic resonance imaging (fMRI) features for enhanced brain connectivity analysis.
• To address the age- and sex-related complexities in ASD diagnosis through a multi-task learning framework.

Main Methods:

• Developed a Sparse-MVTC (Sparse Multi-View Task-Centralized) classification framework.
• Extracted multi-view features from fMRI data to represent brain connectivity.
• Formulated the problem as a multi-view multi-task sparse learning challenge, solved via a task-centralized strategy.
• Partitioned subjects into age/sex-specific groups, treating each as a distinct classification task.

Main Results:

• The Sparse-MVTC method demonstrated superior performance in ASD diagnosis compared to existing methods.
• Experiments conducted on the ABIDE database validated the effectiveness of the proposed approach.
• The task-centralized strategy proved to be a highly efficient solution for the learning problem.

Conclusions:

• The proposed Sparse-MVTC method significantly improves the accuracy of computer-assisted ASD diagnosis.
• Integrating multi-view fMRI data within a multi-task learning framework is effective for ASD detection.
• This approach offers a promising direction for early and accurate diagnosis of autism spectrum disorder.