Deep Learning-based Decision-tree Classifier for Tuberculosis Diagnosis

In recent years, medical disease-assisted diagnosis has been increasingly used. Prior to the COVID-19 epidemic, tuberculosis was the leading cause of death in the single infectious disease that dominated the global epidemic, and approximately 40% of tuberculosis patients were undiagnosed. Thus, making the development of a low-cost, non-invasive digital screening tool important for improving diagnosis in this area. In this paper, based on clinical and demographic data from 1105 patients collected from clinics in seven countries, and cough records from 1082 of these patients combined with convolutional neural networks and light gradient boosting machine to construct a model for the diagnosis of tuberculosis, with the final model achieving an AUC of 0.792 on the test set. This model is therefore a good reference for the auxiliary diagnosis of tuberculosis.