Machine learning for accurate detection of small airway dysfunction-related respiratory changes: an observational...

Area of Science:

• Pulmonary Medicine
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Small airway dysfunction (SAD) diagnosis is challenging in patients with chronic respiratory symptoms and preserved pulmonary function (PPF).
• Machine learning (ML) offers potential for improving SAD diagnosis using impulse oscillometry (IOS) data.
• This study evaluates ML algorithms for SAD detection with IOS analysis.

Purpose of the Study:

• To assess the performance of various ML algorithms in diagnosing SAD using IOS parameters.
• To identify the optimal ML configuration for SAD diagnosis in individuals with PPF.
• To compare ML-based IOS analysis with traditional oscillometric parameters.

Main Methods:

• 280 subjects underwent IOS and spirometry, categorized into healthy controls, normal spirometry, and abnormal spirometry groups.
• Supervised ML algorithms including SVM, RF, ADABOOST, BAYES, and KNN were evaluated.
• Feature selection strategies were applied to identify the most effective IOS parameters for SAD detection.

Main Results:

• The best oscillometric parameter (BOP) showed limited diagnostic value (AUC 0.642-0.769).
• ML algorithms, particularly Random Forests (RF) and Adaptive Boosting (ADABOOST), significantly outperformed BOP (AUC 0.914-0.971).
• ADABOOST performance was robust across feature selection methods, while other classifiers showed marginal improvement.

Conclusions:

• Impulse oscillometry (IOS) combined with ML algorithms presents a novel approach for diagnosing SAD in patients with PPF.
• This combination facilitates earlier detection of respiratory abnormalities in patients with chronic respiratory symptoms.
• The findings support the clinical utility of ML-enhanced IOS for early SAD diagnosis.