Classification of Autism Spectrum Disorder Using Random Support Vector Machine Cluster.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Autism spectrum disorder (ASD) is a neurological developmental disorder characterized by communication and social interaction deficits.
• Machine learning, particularly support vector machines (SVM), is commonly used for classifying ASD patients but often yields low accuracy with single classifiers.
• Existing methods struggle with the complexity of ASD classification, necessitating more robust approaches.

Purpose of the Study:

• To develop and evaluate a novel machine learning method for improved classification of autism spectrum disorder (ASD) using neuroimaging data.
• To enhance the diagnostic accuracy of ASD by employing multiple SVMs in a clustered approach.
• To identify specific brain regions associated with ASD through feature analysis.

Main Methods:

• Utilized resting-state functional magnetic resonance imaging (fMRI) data from the Autism Brain Imaging Data Exchange (ABIDE) database.
• Applied a random support vector machine (SVM) cluster algorithm to classify 61 ASD patients and 46 typical controls (TC).
• Selected a subset of 84 subjects after data preprocessing to exclude those with excessive head motion.

Main Results:

• The random SVM cluster method demonstrated excellent classification performance for ASD.
• Achieved a high classification accuracy of 96.15% using an optimal feature set.
• Identified abnormal brain regions in ASD patients, including the inferior frontal gyrus (IFG), hippocampus, and precuneus.

Conclusions:

• The proposed random SVM cluster method offers a significant improvement in classifying ASD compared to single SVM approaches.
• This machine learning technique shows potential as an auxiliary diagnostic tool for autism spectrum disorder.
• The identified brain regions provide further insights into the neurobiological underpinnings of ASD.