Classification of obstructive and non-obstructive pulmonary diseases on the basis of spirometry using machine learning techniques

摘要

Background: The symptomatic similarities between the two categories of pulmonary diseases, obstructive and non-obstructive, make the early diagnosis difficult for clinicians. Spimmetry is a popular lung investigation that is performed in the early diagnostic stages to understand the mechanics of lungs. This work aims to develop machine learning models to classify obstructive and non-obstructive pulmonary diseases on the basis of spimmetry data. Method: Supervised learning models were developed with support vector machine (SVM), random forest (RF), Naive Bayes (NB) and multi-layer perceptron (MLP) algorithms. Models were trained with spirometry data of 1163 patients using 5-fold cross validation (CV) and further validated with a blind dataset of 151 patients for external validation. Results: The MLP model performed optimally with an accuracy of 83.7% and Matthew's correlation coefficient of 0.682 with 5-fold CV. All the models performed well while validating the blind dataset. The disease-specific prediction of COPD and DPLD, as obstructive and non-obstructive respectively, achieved similar to 90% accuracy in the training dataset. The MLP model was stored in a web server for use in a web application. Conclusions: The machine learning models were able to predict obstructive and non-obstructive pulmonary diseases with good accuracy, based on spirometry data. The web application can be used by clinicians and patients as a tool for early prediction.