Chest Radiography of Tuberculosis: Determination of Activity using Deep Learning Algorithm.

Researchers

Journal

Modalities

Models

Abstract

Inactive or old healed tuberculosis (TB) on chest radiograph (CR) is often found in high TB incidence countries, and differentiation from active TB is important to avoid unnecessary evaluation and medication. This study develops a deep learning (DL) model to estimate activity in a single chest radiographic analysis.A total of 3824 active TB CRs from 511 individuals and 2277 inactive TB CRs from 558 individuals were retrospectively collected. A pretrained convolutional neural network was fine-tuned to classify active and inactive TB. Pretraining was done with 8,964 pneumonia and 8,525 normal cases from the National Institute of Health (NIH) dataset. The DL model learns the following tasks during the pretraining phase: pneumonia vs normal, pneumonia vs active TB, and active TB vs normal. The performance of the DL model was validated using 3 external datasets. Receiver-operating characteristic (ROC) analyses were performed to evaluate diagnostic performance for determination of active TB in DL model and radiologists. Sensitivities and specificities for determination of active TB were evaluated in both DL model and radiologists.The performance of the DL model showed area under the curve (AUC) values of 0.980 in internal validation and 0.815 and 0.887 in external validation. The AUC values for the DL model, thoracic radiologist and general radiologist evaluated using one of the external validation datasets were 0.815, 0.871 and 0.811, respectively.This DL-based algorithm showed its potential as an effective diagnostic tool for identifying TB activity and could be useful for follow-up of patients with inactive TB in high TB burden countries.