Abstract

Background: Inactive or old, healed tuberculosis (TB) on chest radiograph (CR) is oftenfound in high TB incidence countries, and to avoid unnecessary evaluation and medication,differentiation from active TB is important. This study develops a deep learning (DL)model to estimate activity in a single chest radiographic analysis. Methods: A total of 3,824 active TB CRs from 511 individuals and 2,277 inactive TB CRsfrom 558 individuals were retrospectively collected. A pretrained convolutional neuralnetwork was fine-tuned to classify active and inactive TB. The model was pretrainedwith 8,964 pneumonia and 8,525 normal cases from the National Institute of Health(NIH) dataset. During the pretraining phase, the DL model learns the following tasks:pneumonia vs. normal, pneumonia vs. active TB, and active TB vs. normal. The performanceof the DL model was validated using three external datasets. Receiver operatingcharacteristic analyses were performed to evaluate the diagnostic performance to determineactive TB by DL model and radiologists. Sensitivities and specificities for determiningactive TB were evaluated for both the DL model and radiologists. Results: The performance of the DL model showed area under the curve (AUC) valuesof 0.980 in internal validation, and 0.815 and 0.887 in external validation. The AUC valuesfor the DL model, thoracic radiologist, and general radiologist, evaluated using oneof the external validation datasets, were 0.815, 0.871, and 0.811, respectively. Conclusion: This DL-based algorithm showed potential as an effective diagnostic toolto identify TB activity, and could be useful for the follow-up of patients with inactive TBin high TB burden countries.