Evaluation of different decision tree-based methods applied to assessment of bronchial blocking success in patients with destructive forms of tuberculosis

Abstract

We have studied the importance of various lung characteristics obtained by computed tomography (CT), in combination with other factors, for the outcome of endobronchial valve (BV) therapy in patients with destructive lung pathology due to tuberculosis. We have developed predictive models based on the decision tree method and the modern efficient CatBoost algorithm trained on clinical data. This allowed us to identify key characteristics and interactions between them, as well as evaluate the success of endobronchial valve treatment.The constructed models show that the main influence on the positive result of bronchoblocking was provided by non–specific factors (patient’s age and MBT sensitivity) not related to the technique itself.