Performance analysis of artificial intelligence-based classification models for diagnosing asthma in children

Area of Science:

• Pediatric Pulmonology
• Medical Informatics
• Computational Medicine

Background:

• Asthma is a prevalent respiratory condition in children, characterized by symptoms like cough and wheezing.
• Accurate diagnosis of pediatric asthma is crucial for timely and effective management.
• This study investigates the utility of artificial intelligence in enhancing diagnostic precision for childhood asthma.

Purpose of the Study:

• To evaluate the effectiveness of various machine learning models in diagnosing pediatric asthma.
• To compare the performance of different algorithms in distinguishing asthma from non-asthmatic chronic cough in children.
• To identify key clinical predictors for pediatric asthma using data-driven approaches.

Main Methods:

• A cohort of 900 children (aged 6-18 years) with chronic cough was analyzed.
• Eight machine learning models were applied to demographic, clinical, and pulmonary function data.
• Model performance was assessed using metrics including F1 score and ROC AUC.

Main Results:

• Gradient Boosting, Random Forest, and AdaBoost models exhibited high diagnostic performance (F1 scores > 0.969, ROC AUC > 0.995).
• Exercise-induced cough and recurrent bronchiolitis were identified as significant asthma predictors.
• k-Nearest Neighbors showed the lowest accuracy, highlighting variability in model effectiveness.

Conclusions:

• Machine learning algorithms demonstrate substantial potential for improving the accuracy and efficiency of pediatric asthma diagnosis.
• The findings suggest that AI-powered tools can aid clinicians in diagnosing childhood asthma more effectively.
• Further research is warranted to validate and implement these AI models in clinical practice.