Can lung airway geometry be used to predict autism? A preliminary machine learning-based study

Area of Science:

• Medical Imaging
• Biomarkers
• Neurodevelopmental Disorders

Background:

• Autism spectrum disorder (ASD) diagnosis relies on behavioral assessments.
• Objective biomarkers for ASD are needed to aid early detection and understanding.
• Airway geometry has not been extensively studied as a potential biomarker for ASD.

Purpose of the Study:

• To investigate the feasibility of using airway geometry as a biomarker for autism spectrum disorder (ASD).
• To determine if measurable differences in airway branching angles exist between children with ASD and neurotypical controls.

Main Methods:

• Retrospective analysis of chest computed tomography (CT) scans from 31 children with ASD and 23 healthy controls.
• Utilized principal component analysis and support vector machine for feature selection and classification.
• Identified eight key airway branching angles as features for classification.

Main Results:

• Achieved a peak cross-validation accuracy of nearly 89% in classifying ASD cases.
• Demonstrated high sensitivity (94%) but moderate specificity (78%) in distinguishing between groups.
• Identified significant differences in airway branching angles between ASD and control groups.

Conclusions:

• Airway geometry, specifically branching angles, shows potential as a measurable biomarker for ASD.
• The findings suggest a distinct airway phenotype in children with ASD.
• Further research is warranted to refine specificity and clinical applicability.