Automatic classification of pulmonary tuberculosis and sarcoidosis based on random forest

With the accumulation of medical data and rapid development of artificial intelligence, machine learning has entered the medical field, and especially has been widely adopted in disease diagnosis. The essence of disease identification is classification. In this article, we used the medical data of hospitalized patients in our hospital to train random forest classifiers to make disease differentiation between pulmonary tuberculosis and sarcoidosis. Since there were various medical data formats for patients, and these data spreaded in many isolated medical systems, feature selection was difficult for disease classification. We made feature selection automatically only on laboratory result based on some strategies. Using the laboratory result data set, we performed classification with an average AUC of 81% automatically without doctor's intervention. The results of random forest model gave the importance score of each feature, which provided a basis for early diagnosis and optimization of diagnostic processes of pulmonary tuberculosis and sarcoidosis.