Machine learning-based model assists in differentiating Mycobacterium avium Complex Pulmonary Disease from Pulmonary Tuberculosis: A Multicenter Study.

Abstract

The number of Mycobacterium avium-intracellulare complex pulmonary disease patients is increasing globally. Distinguishing Mycobacterium avium-intracellulare complex pulmonary disease from pulmonary tuberculosis is difficult due to similar manifestations and characteristics. We aimed to build and validate a machine learning model using clinical data and computed tomography features to differentiate them. This multi-centered, retrospective study included 169 patients diagnosed with Mycobacterium avium-intracellulare complex and pulmonary tuberculosis from date to date. Data were analyzed, and logistic regression, random forest, and support vector machine models were established and validated. Performance was evaluated using receiver operating characteristic and precision-recall curves. In total, 84 patients with Mycobacterium avium-intracellulare complex pulmonary disease and 85 with pulmonary tuberculosis were analyzed. Patients with Mycobacterium avium-intracellulare complex pulmonary disease were older. Hemoptysis rate, cavity number and morphology, bronchiectasis type, and distribution differed. The support vector machine model performed better. In the training set, the area under the curve was 0.960, and in the validation set it was 0.885. The precision-recall curve showed high accuracy and low recall for the support vector machine model. The support vector machine learning-based model, which integrates clinical data and computed tomography imaging features, exhibited excellent diagnostic performance and can assist in differentiating Mycobacterium avium-intracellulare complex pulmonary disease from pulmonary tuberculosis.