The Development of a Practical Artificial Intelligence Tool for Diagnosing and Evaluating Autism Spectrum Disorder:...

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Biomarker Discovery

Background:

• Autism spectrum disorder (ASD) is a complex neurodevelopmental disorder with unknown etiology.
• Early diagnosis and intervention are crucial for improving ASD patient outcomes.
• Structural magnetic resonance imaging (sMRI) is underutilized for psychiatric disorders like ASD due to subtle anatomical changes.

Purpose of the Study:

• To investigate the potential of identifying structural brain patterns as biomarkers for ASD diagnosis and evaluation.
• To explore the application of AI in detecting subtle neuroanatomical differences in ASD.

Main Methods:

• Developed a novel 2-level histogram-based morphometry (HBM) classification framework.
• Utilized a 3D histogram of oriented gradients (HOG) algorithm for feature extraction from sMRI data.
• Applied the framework to four Autism Brain Imaging Data Exchange (ABIDE) datasets using stratified 10-fold cross-validation and Naive Bayes classification.

Main Results:

• The HBM framework with 3D HOG achieved an Area Under the Curve (AUC) >0.75 across all datasets, with a maximum AUC of 0.849.
• The 3D HOG algorithm demonstrated significant accuracy improvements over 2D HOG (>4%) and Scale-Invariant Feature Transform (SIFT) (>18%).
• Identified known and novel ASD-related brain regions, including the frontal gyrus, temporal gyrus, and hippocampus.

Conclusions:

• Neuroimaging biomarkers for distinguishing ASD patients from healthy controls can be identified using cost-effective sMRI.
• Data-driven AI technology shows significant potential for clinical application in diagnosing neurological and psychiatric disorders with subtle brain changes.