Machine learning-enhanced HRCT analysis for diagnosis and severity assessment in pediatric asthma

Area of Science:

• Pediatric Pulmonology
• Medical Imaging Analysis
• Artificial Intelligence in Medicine

Background:

• Chest high-resolution computed tomography (HRCT) is conditionally recommended for diagnosing conditions mimicking or coexisting with severe asthma in children.
• HRCT scans can reveal structural airway changes in pediatric patients, offering insights beyond standard diagnostic methods.
• Identifying specific imaging biomarkers is crucial for accurate diagnosis and management of severe pediatric asthma.

Purpose of the Study:

• To develop a machine learning (ML)-based model for analyzing chest HRCT images.
• To assist pediatric pulmonologists in identifying features indicative of severe asthma in children.
• To differentiate structural airway abnormalities associated with severe asthma from those in healthy children.

Main Methods:

• A retrospective case-control study comparing children with severe asthma to age- and sex-matched controls.
• Chest HRCT scans were analyzed using statistical methods including classification trees and random forests.
• Conventional ROC analysis was employed to identify significant imaging features differentiating severe asthma.

Main Results:

• Chest HRCT scans effectively differentiated children with severe asthma from controls.
• Significant differences observed in bronchial thickening, airway wall thickness percentage (AWT%), bronchiectasis grading and severity, mucus plugging, and centrilobular emphysema.
• An AWT% of ≥ 38.6 was identified as an optimal classifier for severe asthma, achieving 95% sensitivity, specificity, and accuracy.

Conclusions:

• ML-based analysis of chest HRCT scans shows significant potential for accurately identifying severe asthma features in children.
• This AI-driven diagnostic tool can enhance the evaluation of pediatric asthma.
• The findings support the development of more targeted treatment strategies for severe pediatric asthma based on imaging biomarkers.