High-accuracy detection of airway obstruction in asthma using machine learning algorithms and forced oscillation measurements.

Objective: The main pathologic feature of asthma is episodic airway obstruction. This is usually detected by spirometry and body plethysmography. These tests, however, require a high degree of collaboration and maximal effort on the part of the patient. There is agreement in the literature that there is a demand of research into new technologies to improve non-invasive testing of lung function. The purpose of this study was to develop automatic classifiers to simplify the clinical use and to increase the accuracy of the forced oscillation technique (FOT) in the diagnosis of airway obstruction in patients with asthma.

Methods: The data consisted of FOT parameters obtained from 75 volunteers (39 with obstruction and 36 without). Different supervised machine learning (ML) techniques were investigated, including k-nearest neighbors (KNN), random forest (RF), AdaBoost with decision trees (ADAB) and feature-based dissimilarity space classifier (FDSC).

Results: The first part of this study showed that the best FOT parameter was the resonance frequency (AUC = 0.81), which indicates moderate accuracy (0.70-0.90). In the second part of this study, the use of the cited ML techniques was investigated. All the classifiers improved the diagnostic accuracy. Notably, ADAB and KNN were very close to achieving high accuracy (AUC = 0.88 and 0.89, respectively). Experiments including the cross products of the FOT parameters showed that all the classifiers improved the diagnosis accuracy and KNN was able to reach a higher accuracy range (AUC = 0.91).

Conclusions: Machine learning classifiers can help in the diagnosis of airway obstruction in asthma patients, and they can assist clinicians in airway obstruction identification.