Protein data in the identification and stage prediction of bronchopulmonary dysplasia on preterm infants: a machine learning study

Abstract

Bronchopulmonary Dysplasia (BPD) is a chronic lung disease mostly affecting the premature newborns who are in the need for oxygen therapy. Main reason for this disease is underdeveloped lungs which need the help of ventilator to expand and breath. This is a very serious disease without a specific test for diagnosis. As treatment is very important for this disease in order to improve the lung function of the baby, on time diagnosis is crucial. Hence, this study checks the potential of protein data in the diagnosis of BPD, and also in the prediction of BPD stage. Mutual information is used in the selection of relevant features of each study. Selected set of features are used with different machine learning algorithms and the accuracies among the models are compared. By this comparison, this study reveals the best number of features in each of the prediction along with the best machine learning algorithm. The highest accuracy value obtained in the diagnosis model shows that it can be used in practice to improve the diagnosis accuracy. However, it seems that prediction of the disease stage is a far complex problem which needs further improvement in its accuracy.