A deep learning model using chest X-ray for identifying TB and NTM-LD patients: a cross-sectional study

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Pulmonary Medicine

Background:

• Pulmonary tuberculosis (TB) and nontuberculous mycobacterial lung disease (NTM-LD) present similar radiographic findings, complicating diagnosis.
• Accurate differentiation is crucial due to differing infectiousness and treatment protocols for TB and NTM-LD.

Purpose of the Study:

• To evaluate the efficacy of an artificial intelligence (AI) deep neural network (DNN) in distinguishing between pulmonary TB and NTM-LD using chest X-rays (CXRs).
• To compare the AI model's diagnostic performance against that of human pulmonologists.

Main Methods:

• A dataset of 1500 CXRs from patients with pulmonary TB, NTM-LD, and controls (Imitators) was retrospectively analyzed.
• A deep neural network (DNN) was developed and its performance assessed using area under the receiver operating characteristic curves (AUC) on internal and external test sets.
• A reader study compared the DNN model's accuracy against senior and junior pulmonologists under varying prevalence scenarios.

Main Results:

• The DNN model achieved AUCs of 0.83 (TB) and 0.86 (NTM-LD) on the internal test set, and 0.76 (TB) and 0.64 (NTM-LD) on the external set.
• The AI model demonstrated significantly higher classification accuracy (66.5%) compared to senior (50.8%) and junior (47.5%) pulmonologists.
• The DNN model maintained stable performance across different mycobacterial prevalence scenarios.

Conclusions:

• The developed DNN model exhibits satisfactory performance in classifying patients with presumptive mycobacterial lung diseases.
• The AI model's accuracy surpasses that of human pulmonologists, suggesting its potential as a complementary first-line screening tool.
• AI-powered CXR analysis offers a promising avenue for improving the early and accurate diagnosis of mycobacterial lung diseases.